SlideShare a Scribd company logo
1 of 28
Download to read offline
A folksonomy-based lightweight resource annotation metadata schema
for personalized hypermedia learning resource delivery
Simon Boung-Yew Laua
, Chien-Sing Leeb
* and Yashwant Prasad Singhc
a
Faculty of Engineering and Science, Universiti Tunku Abdul Rahman, 53300 Setapak, Kuala
Lumpur, Malaysia; b
Graduate Institute of Network Learning Technology, National Central
University, 300, Jhongda Road, Jhongli City, Taoyuan 32001, Taiwan; c
Faculty of Information
Technology, Multimedia University, Jalan Multimedia, 63100 Cyberjaya, Malaysia
(Received 1 December 2011; ļ¬nal version received 12 August 2012)
With the proliferation of social Web applications, users can now collaboratively
author, share and access hypermedia learning resources, contributing to richer
learning experiences outside formal education. These resources may or may not
be educational. However, they can be harnessed for educational purposes by
adapting and personalizing them to diļ¬€erent learner needs. We propose to
collaboratively annotate learning resources with a lightweight resource annota-
tion metadata schema complemented by a folksonomy-derived semantic model.
The annotation metadata schema follows a novel policy to associate numerical
ratings to learnersā€™ subjective expression of opinions on learning resources. On the
other hand, the semantic model serves to support a collaborative resource
recommendation algorithm based on the k-nearest neighbour approach. Proof-of-
concept is demonstrated via a prototype Web-based recommender. Preliminary
results indicate learners are conļ¬dent with the recommended learning resources in
terms of accuracy, usefulness, novelty and information adequacy. Formalization
of folksonomy in annotating learning resources as well as in opinion mining from
the crowd to rate resources may not only support personalization but also engage
learners more eļ¬€ectively in technology-enhanced learning.
Keywords: social Web; hypermedia learning resources; personalization;
folksonomy; resource annotation metadata; collaborative ļ¬ltering; technology-
enhanced learning
1. Introduction and motivation
The nature of conventional education is top-down where learning resources are
mainly prepared and authored by domain experts and instructors. Knowledge is
delivered from the instructors to learners and takes the approach of ā€˜ā€˜one size ļ¬ts
allā€™ā€™. This research is motivated by a hypothetical learning scenario illustrated as
follows:
A young adult, Paul in his early 20s does not like to read academic textbooks or attend
training classes provided by experts. He would like to self-learn a highly technical topic
such as ā€˜ā€˜Personal Financeā€™ā€™ (ļ¬nance) or ā€˜ā€˜Mobile and Satellite Communicationā€™ā€™
*Corresponding author. Email: cslee@cl.ncu.edu.tw
Ɠ 2012 Taylor & Francis
Interactive Learning Environments, 2015
Vol. 23, No. 1, 79ā€“105, http://dx.doi.org/10.1080/10494820.2012.745429
(engineering) new to him without going to school or without the guidance of any
instructor. As he has mobile broadband access at his ļ¬ngertips, he intends to source for
useful educational resources suitable for a beginner like him on the Internet. He is on
social network such as Facebook almost daily. He reckons that he can gain access to
relevant learning resources anytime, anywhere and with the help of anyone who shares
and tags various Web articles, blog posts, videos, audio podcasts etc. on a social
network. Hence, such tagging and sharing activities may be of help to Paul provided he
knows where they are (they are easily searchable). Similarly, other users may beneļ¬t
from what Paul tags and shares through his daily Web activities.
As reported in Shroļ¬€ (2011), the ā€˜ā€˜Y generationā€™ā€™ or ā€˜ā€˜Facebook Generationā€™ā€™
spends an average 5 h a day on social network and Web 2.0 applications. The advent
and prevalence of Web 2.0 or ā€˜ā€˜Read-Writeā€™ā€™ Web technologies such as social
network, blogs, wikis and social bookmarks among the younger generation has
facilitated social interaction and collaboration among users and, hence, opened the
door for authoring and sharing of educational hypermedia resources such as Web
articles, videos, images and podcasts for personalized self-paced learning on the Web
(Ulbrich, Kandpal, & Tochtermann, 2003).
Although the increase in the number of resources available in the open Web can
be harnessed for educational purposes, they may or may not be ļ¬t for educational
purposes for all users. What is evidently necessary for this mode of learning is not
only a facility to understand the educational needs of learners, but also a tool to
enable learning resources to be annotated for personalization of these resources to
diļ¬€erent learner needs.
Due to the overwhelmingly huge amount of learning resources generated and
available online, we argue that annotation of such resources by a small group of
instructional designers or domain experts is not only time-consuming, but also
prone to be inaccurate (due to the lack of time interacting with every piece of
information and timely updates). We thus propose that user participation in the
annotation process would not only reduce the cost of updating and maintaining
resource description metadata, but also improve the quality of resource
description such as serendipity (Mathes, 2004; Quintarelli, 2005) and better
generalization of new knowledge (Swarup & Gasser, 2008) from a larger pool of
diverse opinions.
To enable resource discovery and adaptation for personalized learning,
collaborative ļ¬ltering (CF) is commonly adopted to label, classify, select and ļ¬lter
hypermedia resources on the Web. The CF based on attribute description or
numerical ratings to enable resource discovery and adaptation for personalized
learning is relatively new (Khribi, Jemni, & Nasraoui, 2007, 2009; Itmazi & Megias,
2008; Tan, Guo, & Li, 2008).
In the latter type of CF, it builds a model from a community of usersā€™ past
behaviour (such as ratings given to items) to predict ratings for items that the user
may have an interest in. However, generally, contextual information such as ā€˜ā€˜whoā€™ā€™
and ā€˜ā€˜whyā€™ā€™ a resource is being liked or disliked is not suļ¬ƒciently captured in CF. In
addition, each user may perceive and rate a resource from diļ¬€erent perspectives. For
instance, a video tutorial which is ranked 5 on a scale of 1 (worst) to 5 (best) by one
user may be due to the quality of its content, while the resource may be ranked 5 by
another user due to his preference to the way the content is being presented. Both
users may also diļ¬€er signiļ¬cantly in the way they learn.
To close the gap, we look towards using collaborative tagging to describe
learning resources.
S.B.-Y. Lau et al.
80
Collaborative tagging is the collective action of users associating tags to resources
created or experienced by them (Bouillet, Feblowitz, Liu, Ranganathan, & Riabov,
2008; Pfeiļ¬€er & Tonkin, 2009). The classiļ¬cation of these resources using tags
produces folksonomy. Folksonomy is spontaneous, non-hierarchical categorization
of Web resources using shared keywords by a community of users (Li & Lu, 2008;
Vanderwal, 2007).
Conceptually, a generic model of collaborative tagging comprises a tripartite
three-uniform structure of entities, i.e. the users, tags and resources. Hence, a
collaborative tagging system can be represented as a tripartite hypergraph with
hyperedges. Each hyperedge will have vertices of three types which represent a user,
a resource being tagged and a tag in a folksonomy where the set of vertices on the
folksonomy can be partitioned into three disjoint sets of users, U Ā¼ {u1, u2, . . . , uk},
tags, T Ā¼ {t1, t2, . . . , tl} and resources, R Ā¼ {r1, r2, . . . , rm}. A folksonomy network
can be visualized as a tripartite hypergraph
H Ā¼ V; E
ư ƞ ư1ƞ
where V denotes the vertices in the tripartite graph, V Ā¼ U [ T [ R and E denotes
the set of hyperedges. Each hyperedge connects exactly three vertices, one from
each set of U, T and R where E  U 6 T 6 R Ā¼ {(u,t,r) j u 2 U, t 2 T, r 2 R)} is a
ternary relation among U, T and R whose elements are instances of annotation or
tag assignment by a particular user from U with a tag in T for a resource in R
(Mika, 2005; Zhang  Chuang, 2010). Hence, folksonomy has great potential in
enriching learning resource description metadata as well as automated selection,
ļ¬ltering and ranking of learning resources on the open Web for personalization of
these resources to learnersā€™ educational needs. To the best of our knowledge, there
is a lack of semantic support for social tags for learning in social contexts. Thus,
the research problem addressed in this article is how to enable the learning system
to make sense of the free, unconstrained and arbitrary tag metadata (Guy 
Tonkin, 2006). Our work is centred on the research question: How can we
formalize the deļ¬nition of user-contributed tag metadata to facilitate
personalization?
We propose a semantic model for folksonomy-derived resource description
metadata and a policy for resource annotation based on the semantic model to
formalize semantic information about learning resources as well as users who tag
them. This enables resources to be selected based on user context proļ¬le (e.g.
learning style and competency level) and educational preferences (e.g. theme, content
format and complexity level of resources). The working principle of how the
description metadata schema can be integrated into a learning resource recommen-
dation approach based on k-nearest neighbour will be presented in Section 8.
2. Contribution
The focus of our work is diļ¬€erent from the above-mentioned related work on
folksonomy from two perspectives. Firstly, our work focuses on the application of
learning resource personalization in informal, self-directed social learning environ-
ment while most related work focus on conventional e-learning application in formal
education settings.
Interactive Learning Environments 81
Secondly, as far as CF of learning resources is concerned, there is currently
limited formal conceptualization that represents folksonomy in a consistent way
(Kim, Breslin, Yang,  Kim, 2008) Consistent representation of folksonomy
facilitates systematic annotation process. Furthermore, there is no semantic
annotation schema representing knowledge about learning resources attained from
tags. Hence, we aim to formulate a novel semantic model for tags which serves as a
tool and policy framework to associate and correlate usersā€™ subjective tag keywords
metadata to high-level concepts depicting features of a learning resource such as
theme, content format and complexity level as well as numerical rating of the
resource. The model includes a well-deļ¬ned resource rating policy based on user
opinion. The semantic model and annotation technique address the issue of
formalizing collaborative social tags in informal learning. Besides, a novel domain
ontology for ā€˜ā€˜personal ļ¬nanceā€™ā€™ was devised for experimentation because as far as
we have surveyed, there is currently no readily deļ¬ned ontology for this domain.
Thirdly, another important aspect of learning resource description proļ¬ling is
mining the opinions of the user community about learning resources. At present, two
popular opinion mining techniques on Web 2.0 systems are the ā€˜ā€˜Likeā€™ā€™ voting on the
Facebook social network platform and numerical ratings. These techniques generally
suļ¬€er from a weakness: They do not naturally take into consideration ā€˜ā€˜contextā€™ā€™ in
which an item is being rated (Nakamoto, Nakajima, Miyazaki,  Uemura, 2008).
Hence, we also contribute to ļ¬ll this gap by devising a user ā€˜ā€˜opinion-ratingā€™ā€™ policy.
This policy is devised to extract the feelings of users about how ā€˜ā€˜goodā€™ā€™ a learning
resource is for the purpose of rating and ranking the resources.
Lastly, our semantic model provides the framework for
(1) item-based ļ¬ltering of learning resources prior to a user-based classiļ¬cation
of resources and
(2) ranking of learning resources based on numerical ratings inferred from tags
within a collaborative resource recommendation algorithm based on the k-
nearest neighbour.
3. Related work
Learning resource recommendation is at a relatively early stage of research and
development (Itmazi  Megias, 2008). A major concern among the research
fraternity is identifying the requirements of learning recommender systems
(Drachsler, Hummel,  Koper, 2007; Santos, Baldiris, Boticario1, Restrepo1, 
Fabregat, 2011; Santos  Boticario, 2011) with issues and challenges speciļ¬c to the
learning domain (Tang  McCalla, 2005). Most current eļ¬€orts on recommender
systems are related to the formation of conceptual models, frameworks or
architecture (Itmazi  Gea; 2006; Itmazi  Megias, 2008; Kerkiri, Manitsaris, 
Mavridou, 2007; Otair  Al Hamad, 2005; Tan et al., 2008) focusing on formal
learning (Karampiperis  Sampson, 2005). The designs are, in essence, top-down
and highly structured where learning material and learning path are designed and
administered by domain experts to guarantee quality. However, implementation
details of these eļ¬€orts are not in abundance for conclusive remarks to be made about
the level of success of such applications.
Meanwhile, learning recommender systems for informal learning are slowly
developing (Drachsler  Manouselis, 2009; Soonthornphisaj, Rojsattarat, 
S.B.-Y. Lau et al.
82
Yim-ngam, 2006). However, there is limited documented information about the
implementation detail and the level of success of these systems. Overview of the
requirements for learning personalization and the types of recommendation
approaches are summarized in Figure 1.
As far as learning resource description proļ¬ling is concerned, Fok and Ip (2007)
designed and constructed personalized education ontology (PEOnto) through an
iterative ontology construction process. The PEOnto is made up of ļ¬ve ontologies of
vocabularies. These are: the users, the learning content, the curriculum, the pedagogy
and the software agents for describing, organizing, retrieving and recommending
hypermedia educational resources. Fok and Ip (2007) have since adopted PEOnto in
a prototype called personalized instruction planner (PIP). The PIP is implemented to
facilitate instructors in annotating learning objects with semantic description.
Meanwhile, Jovanovic, GasĢŒevicĢ, and DevedzĢŒicĢ (2009) devised an ontological
framework for learning personalization, which includes a learning content structural
ontology, a context ontology to specify the pedagogical role of the content, a domain
ontology of concepts, a learning path ontology and a user ontology. These
ontologies are used to personalize learning content to the usersā€™ domain knowledge,
preferences and learning styles in the domain of intelligent information systems.
Both research eļ¬€orts have demonstrated the use of ontologies in authoring learning
resources in formal education.
Other related work on ontology-based recommendation systems driven by
similar motivation (Buriano et al., 2006; Cantador, BellogıĢn,  Castells, 2008; Costa,
Guizzardi, Guizzardi,  Filho, 2007; Naudet, Mignon, Lecaque, Hazotte,  Groues,
2008) have endeavoured to enhance the semantic description of user preferences and
items with concepts and instances deļ¬ned in ontologies. Eļ¬€orts to construct
ontologies to represent knowledge and concepts in e-learning include those by
Heiyanthuduwage and Karunaratne (2006), SĢŒimuĢn, Andrejko, and BielikovaĢ (2007)
and Tsai, Chiu, Lee, and Wang (2006). Some of these eļ¬€orts are summarized in
Table 1. These approaches exploit learning personalization based on domain
knowledge and user context.
Figure 1. Framework for learning resource recommender systems.
Interactive Learning Environments 83
There are various approaches employed in constructing domain ontologies.
Apart from the knowledge engineering approach adopted in this work (Yun, Xu,
Wei,  Xiong, 2009), there are the iterative convergence model (Pinto  Peralta,
2003), composition/integration model (Lin, Tseng, Weng, Lin,  Su, 2007), skeletal
(Uschold  GruĢˆningerm, 1996), seven-step method (Yun et al., 2009) and ļ¬ve-step
recipe (Yun et al., 2009), whose phases of development are conceptually similar. In
this article, we adopt the widely applied knowledge engineering approach which is
generically modelled after the IEEE 1074-2006 Software Development Life Cycle
processes and proven eļ¬€ective in the e-learning domain (Yun et al., 2009) to develop
the ontology for the personal ļ¬nance domain.
More recently, there is growing popularity on leveraging collaborative tags for
context clues to supersede CF based on user ratings (Gadepalli, Rundensteiner,
Brown,  Claypool, 2010; Milicevic, Nanopoulos,  Ivanovic, 2010; Nakamoto
et al., 2008; Tso-Sutter, Marinho,  Schmidt-Thieme, 2008). This initiative is based
on the assumption that similar users will use similar or related keywords to express
their opinions on similar items. This is also the opinion harvesting approach that our
article is presenting. Details of the approach will be presented in Section 6.4.
Meanwhile, resource recommendation approaches can be classiļ¬ed based on the
way ļ¬ltering and selection of resources are done, i.e. content-based (item similarity)
and CF-based (user similarity) or based on the ways in which ratings of the resources
are being estimated, i.e. memory-based and model-based approaches. Hybrid
approaches combine more than one approach for speciļ¬c purposes. Taxonomy of
the resource recommendation approaches is shown in Figure 2.
In general, most recommendation approaches in personalized learning systems
employ the CF approach compared to the content-based approach due to the
presence of the concept of ā€˜ā€˜communityā€™ā€™ in CF, which makes it a natural choice for
learning in a social-networked environment (Khribi et al., 2007, 2009). User-based
CF such as k-nearest neighbour (k-NN) is the most prevalent approach employed for
learning resource recommendation (Itmazi  Megias, 2008; Khribi et al., 2009; Tan
et al., 2008). For instance, Khribi et al. (2009) classify learners using k-NN to
recommend Web links to learners based on cosine similarity. On the other hand,
Shih and Lee (2001) employ the k-NN to provide adaptive learning materials for a
Table 1. Summary of work which exploits ontologies to represent knowledge and concepts.
Semantic technology
exploited Personalization Application
Heiyanthuduwage
and
Karunaratne
(2006)
Ontology that deļ¬nes
relationship between
metadata for learning
objects
Maps user preferences
with learning content
Conventional
open sourced
learning
management
system (LMS)
SĢŒimuĢn et al.
(2007)
Ontology of knowledge
item space and
concept space
Adapt learning
application to user
model
Web-based
e-learning
application
Tsai et al.
(2006)
Ontology of course
content
Rank the degree of
relevance of learning
objects to a userā€™s
preference and nearest
neighboursā€™
preferences
Not stated
S.B.-Y. Lau et al.
84
computer networking course. It is interesting to note that to date, as far as we have
surveyed, item-based CF is yet to be employed prevalently in the e-learning domain
in any signiļ¬cant way.
4. Outline of the article
The rest of this article is organized as follows. Firstly, we conceptualize the
formalization of the meanings of tags in keyword metadata in folksonomy. In Section
6, we present the ontology engineering process of a semantic model for tags. In
Section 7, we demonstrate how the semantic ontology for tags can be applied in the
resource annotation process for a collaborative resource recommendation algorithm
based on k-nearest neighbour approach. In Section 8, we present the user evaluation
results of a prototype Web-based learning recommender system implementing the
said model and algorithm.
5. Formalizing collaborative tags
Conventionally, there is no formal conceptualization for representing folksonomy in a
consistent way (Kim et al., 2008). There is no systematic process to guide the tagging
process and a semantic annotation schema representing the knowledge attained from
tags. As far as we know, there is no commonly known popular Web 2.0 tagging system
representative of the learning scenario we are targeting, which focuses on educational
resources with large enough data size that warrants a conclusive analysis of tags.
As far as we know, currently there is no commonly known popular Web 2.0
tagging system on educational resources that has large enough data size which
warrants a conclusive analysis of tags. Hence, we select Delicious.com as the sample
tag database for our observation. We assume every piece of Web resource can be
educational to speciļ¬c segments of audiences in speciļ¬c contextual situations.
Our study indicates that folksonomy can be principally classiļ¬ed into categories as
follows (Table 2; Marchetti, Tesconi,  Ronzano, 2007; Xu, Fu, Mao,  Su, 2006):
(1) Factual: up to 80% of tags describe the theme of the content being tagged and
approximately 17% describe the content format of the resource or the
context in which the resource is used;
Figure 2. Taxonomy of resource recommendation approaches.
Interactive Learning Environments 85
(2) Descriptive: about 3% of tags describe usersā€™ subjective emotions or opinions.
Though the study stated above clearly indicates the presence of distinct types of
semantic information in tag data, it is inherently diļ¬ƒcult to automatically extract the
implicit meanings of tag data since there is lack of organization or structure of
higher-level semantic relations between tags. A single user ā€˜ā€˜votesā€™ā€™ for a resource
with its tag and the votes are counted only as one in standard tagging applications.
That is, if tag1 Ā¼ tag2, then there is no diļ¬€erence semantically between the ļ¬rst and
subsequent assertions, which means the additional assertions are logically
redundant. For the purpose of harnessing tags to describe learning resources, we
need to devise a policy to guide the tagging process and organize of the meanings of
tags. Simply put, it is to make folksonomy able to ā€˜ā€˜talkā€™ā€™ to a learning system (Li 
Lu, 2008). We propose to undertake this by constructing a semantic model for tags
to address the following semantic information about learning resources:
(1) the theme of the subject matter the learning resource is related to: what the
resource is about (tags such as ā€˜ā€˜personal ļ¬nanceā€™ā€™ and ā€˜ā€˜mobile
communicationā€™ā€™),
(2) the content format of the learning resource (tags such as ā€˜ā€˜videoā€™ā€™ and ā€˜ā€˜textā€™ā€™),
(3) the quality attributes of the resource perceived by users (tags such as
ā€˜ā€˜boringā€™ā€™ and ā€˜ā€˜coolā€™ā€™) and
(4) proļ¬le of the tagger who creates and owns the tag.
We thus propose formalizing folksonomy with a semantic model for tags as
shown in Figure 3 with the following features:
(1) a domain ontology that describes the relations of concepts,
(2) a content format ontology and
(3) an opinion-rating policy to associate numerical rating to subjective opinions
of users to grade the learning resources.
For proof-of-concept, formulation of the semantic model is presented in detail in
the following sections.
6. Semantic model for tags
6.1. Domain ontology
In this section, we present the knowledge engineering process to constructing a novel
domain ontology on ā€˜ā€˜personal ļ¬nanceā€™ā€™ for illustration purpose. The goal of the
Table 2. Breakdown of tags by type of semantic information.
Category Type of semantic information Percentage (%)
Factual Theme of a resource (content-/context-based tags) 78.06
Content format (the kind of informative content) 17.35
Descriptive Subjective opinion or emotion about the level of
ā€˜ā€˜goodnessā€™ā€™
3.06
Others Example: preposition, conjunction and
organizational
1.53
S.B.-Y. Lau et al.
86
domain ontology is to provide the basis for organizing the domain themes. The
purpose of the domain ontology is to guide the learning resource annotation process
to deļ¬ne concepts in the knowledge domain. The domain ontology is extensible with
concepts deļ¬ned by users in the form of tags. It will be implemented in the future
versions of the prototype system described in Section 9.
The knowledge engineering approach to construct the domain ontology is
adopted from the IEEE 1074-2006 Software Development Life Cycle standard
reported in Yun (2009) and Yun et al. (2009). Phases of the ontology construction
process include ontology competence speciļ¬cation, ontology acquisition, ontology
implementation and ontology evaluation as shown in Figure 4.
Firstly, the goal and scope of the ontology, its intended uses and target end users
are deļ¬ned and speciļ¬ed. Competency questions for the ontology are designed
through a few iterations of deliberations and reviews to test whether the goals and
scope of the ontology have been expressively covered. Whether a concept or axiom is
to be included in the domain ontology are veriļ¬ed by the questions. For instance,
competency questions for ā€˜ā€˜fundamental concepts in personal ļ¬nanceā€™ā€™ are such as:
(1) What are the general principles of personal ļ¬nance?
(2) What are the fundamental concepts of personal ļ¬nance for a learner
(beginner) to know?
(3) What are the key knowledge areas related to personal ļ¬nance?
According to the above-mentioned ontology competency questions, relevant
concepts, relations, properties and roles are identiļ¬ed and organized into a semantic
structure by following these steps (Lau  Lee, 2012):
Figure 4. Ontology engineering process of the domain ontology.
Figure 3. Semantic model for tags.
Interactive Learning Environments 87
(1) Enumerate important concepts and terms: important concepts and terms for
the subject domain are manually extracted from related closed-corpus and
open-corpus materials via a careful process of deliberation. Closed-corpus
materials include textbooks and reference books (Altfest, 2007; Bajtelsmit,
2006; Downey, 2005; Frasca, 2009; Kapoor, 2009; Michael, 1997) and online
educational materials (Blecksmith, 2010; Keown, 2010) authored by domain
experts. Open-corpus materials include open Web resources such as wiki or
blog posts such as GetRichSlowly.com, MyMoney.com and Wisebread.com,
news portals (e.g. CNN Money, Forbes Finance and Yahoo Finance) and
DMOZ Open Directory Project (2012) about personal ļ¬nance. The titles,
subtitles, outlines, table of content and so on were observed to extract the
intrinsic semantic structure of the subject domain. Terms or keywords
extracted from these materials form the conceptual model.
Apart from the information extraction approach described above, in recent
years, it may be also possible to learn the ontologies from the Web using
automated tools. However, as far as we know, most of these semantic Web
tools are still at the early stage of implementation and there is rarely
universally constructed and accepted ontologies on domains such as
ā€˜ā€˜personal ļ¬nanceā€™ā€™. An experiment with Swoogle, a semantic Web search
engine shows that there is currently no any readily deļ¬ned ontology which is
near to the targeted domain in the area of ā€˜ā€˜personal ļ¬nanceā€™ā€™.
(2) Deļ¬ne concepts, attributes and relations of concepts and organize hierarchy
structure: Concepts in the domain are categorized and organized top-down
into hierarchical structures as shown in Figure 5.
For this research, the approach proposed by Leo, Ceusters, Mani, Ray, and
Smith (2007) is adopted to verify the ontology where the ontology is assessed by
domain experts against a set of criteria. The purpose of the evaluation is not only to
assess the quality of the ontology, but also to enhance the ontology based on the
suggestions of the experts.
The quality of the domain ontology was evaluated against a set of 12
competency statements by a total of ļ¬ve subject experts in the Finance
Department, Faculty of Management of Multimedia University on a Likert scale
from ā€˜ā€˜strongly disagreeā€™ā€™ (1) to ā€˜ā€˜strongly agreeā€™ā€™ (5). The results of the assessment
are as shown in Table 3.
For the purpose of comparison, the ratings are aggregated into a quality score by
summing the product of rating and the number of evaluators. In conclusion, the
domain ontology passes all the 12 quality criteria by attaining average scores
between 17 and 20 out of 25. Subsequently, the ontology was also validated against
the competency questions by the same group of experts. The domain ontology under
experiment was unanimously rated as fulļ¬lling all competency questions.
6.2. Hypermedia content format ontology
Content format indicates whether the learning resource is a text, image, audio, video,
vendor speciļ¬c ļ¬le format, Web service, etc. The hypermedia content format
ontology is devised with reference to the deļ¬nition of Internet media type (Grafman
Productions, 1997) to guide users to tag learning resources with the right content
format. The tags specifying the content format are used to ļ¬lter learning resources.
S.B.-Y. Lau et al.
88
Figure 5. Sample ontology of fundamental concepts in personal ļ¬nance.
Interactive
Learning
Environments
89
As shown in Figure 6, common content formats for hypermedia resources are text,
image, audio, video and Web applications.
6.3. Complexity level
Apart from the above-mentioned keywords, complexity level of the learning
resources can also be tagged with a set of tag keywords, T Ā¼ {ā€˜ā€˜Not Complexā€™ā€™,
Table 3. Ontology quality evaluation results.
Number of evaluators who rated Score
1 2 3 4 5
No. Criteria
Strongly
disagree Neutral
Strongly
agree Score Average
Std
dev.
1. The ontology suļ¬ƒciently and completely
deļ¬nes all the principal high-level
concepts/terms in personal ļ¬nance.
ā€“ 1 ā€“ 4 ā€“ 18 3.6 2.12
2. The ontology has completely deļ¬ned all
the principal relations in personal
ļ¬nance and used them in consistent
ways.
ā€“ 1 ā€“ 4 ā€“ 18 3.6 2.12
3. The use of vocabularies in the ontology is
consistent and unambiguous.
ā€“ ā€“ ā€“ 5 ā€“ 20 4 ā€“
4. The ontology is correct and does not
require much corrections and/or
additions of terms.
ā€“ 1 ā€“ 4 ā€“ 18 3.6 2.12
5. The ontology is accurate for its intended
use, i.e. for learning and mastering a
wide spectrum of concepts in personal
ļ¬nance.
ā€“ ā€“ 2 3 ā€“ 18 3.6 0.71
6. The ontology is compact and does not
have any redundant deļ¬nition of
vocabularies or ā€˜ā€˜circular deļ¬nitionsā€™ā€™.
ā€“ ā€“ 2 3 ā€“ 18 3.6 0.71
7. The ontology is precise and does not give
a new meaning to a term in place of its
original vocabulary already established
in use in personal ļ¬nance.
ā€“ ā€“ 3 2 ā€“ 17 3.4 0.71
8. The ontology is comprehensive and
contains suļ¬ƒcient high-level concepts
to understand the overall big picture of
ā€˜ā€˜personal ļ¬nanceā€™ā€™.
ā€“ ā€“ 2 3 ā€“ 18 3.6 0.71
9. The ontology is detailed enough to
contain suļ¬ƒcient low-level concepts to
have in-depth and comprehensive study
of personal ļ¬nance.
ā€“ ā€“ 1 4 ā€“ 19 3.8 2.12
10. The ontology contains vocabularies which
are formal and standard in the area of
personal ļ¬nance. There is small or no
informal or ad hoc non-standard terms
used.
ā€“ ā€“ 1 4 ā€“ 19 3.8 2.12
11. The ontology is easily extensible. It is
possible to easily add lower-level
concepts into this ontology and extend
the knowledge domain with its current
structure.
ā€“ ā€“ 2 3 ā€“ 18 3.6 0.71
12. The ontology is up to date. It reļ¬‚ects the
most current and up-to-date view of
the fundamental concepts in personal
ļ¬nance.
ā€“ ā€“ 2 3 ā€“ 18 3.6 0.71
S.B.-Y. Lau et al.
90
ā€˜ā€˜Complexā€™ā€™, ā€˜ā€˜Quite Complexā€™ā€™, ā€˜ā€˜Very Complexā€™ā€™, ā€˜ā€˜Extremely Complexā€™ā€™} which can
be translated into numerical values g(t) as shown in Table 4. For instance, if a user
thinks that a resource is ā€˜ā€˜Not Complexā€™ā€™ at all, he is indirectly assigning a value of
ā€˜ā€˜0ā€™ā€™ to the ā€˜ā€˜Complexityā€™ā€™ ļ¬eld of the resource query.
6.4. Opinion-rating policy
The domain ontology and hypermedia content format ontology are used to
determine the degree of relevance of learning resources to a userā€™s query. They are
not able to capture quality of the resources as perceived by users. To ļ¬ll this gap, a
user ā€˜ā€˜opinion-ratingā€™ā€™ policy is devised to extract the opinions of users about how
ā€˜ā€˜goodā€™ā€™ a learning resource is for the purpose of rating and ranking the resources.
The policy functions to associate commonly expressed subjective opinion keywords
of users appearing in tag data to objective numerical rating scores as a product
function:
Composite rating score; g Ā¼ g d
ư ƞ  g l
ư ƞ  g o
ư ƞ ư2ƞ
Table 4. Quantifying complexity level of resources.
Tag, t Complexity level, g(t)
ā€˜ā€˜Notā€™ā€™ ā€˜ā€˜Complexā€™ā€™ 0
ā€˜ā€˜ā€™ā€™ 1
ā€˜ā€˜Quiteā€™ā€™ 2
ā€˜ā€˜Veryā€™ā€™ 3
ā€˜ā€˜Extremelyā€™ā€™ 4
Figure 6. Content format ontology.
Interactive Learning Environments 91
where d is the member of a set of keywords expressing intensity of feeling Ā¼
{ā€˜ā€˜Slightlyā€™ā€™, ā€˜ā€˜Moderatelyā€™ā€™, ā€˜ā€˜Veryā€™ā€™, ā€˜ā€˜Extremelyā€™ā€™}, l Ā¼ {ā€˜ā€˜ā€“ā€™ā€™, ā€˜ā€˜notā€™ā€™} and o is the
member of a set of opinion keywords such as ā€˜ā€˜coolā€™ā€™, ā€˜ā€˜boringā€™ā€™ and ā€˜ā€˜badā€™ā€™. Each of
the terms stated above are translated into corresponding numerical rating score,
g Ā¼ g(d) 6 g(l) 6 g(o) as shown in Table 5.
For instance, a learning resource tagged as ā€˜ā€˜very boringā€™ā€™ will be rated with
rating score, g Ā¼ 4 6 1 6 (71) Ā¼ 7 4 on a scale of 75 to 5 where ā€˜ā€˜veryā€™ā€™ Ā¼ 4,
ā€˜ā€˜7ā€™ā€™ Ā¼ 1 and ā€˜ā€˜boringā€™ā€™ Ā¼ 71.
The strength of this model is
(1) Users are able to rate the learning resources subjectively with phrases
expressing their feelings without having to think about the corresponding
rating value;
(2) Tags are standardized to a standardized system interpretable format. Rating
values can converge to values that rank resources without regard to the
diverse ways of expressions of feelings by users.
7. Collaborative resource recommendation
In a personalized learning process, context metadata (e.g. learning style and
competency level) of a learner, and the need and intention of a learner (e.g. the
desired theme, content format and complexity level) are elicited indirectly or
speciļ¬ed directly by a learner. The query metadata schema is composed of two
components, namely the learner context proļ¬le and information need proļ¬le. A
query is deļ¬ned as an ordered vector, Q Ā¼ ((Learning Style, Competency Level),
(Theme, Content Format, Complexity)). The resource query is the result of a learner
context proļ¬ling and a resource annotation process. Schema and policy described of
the semantic model for tags described above formalize these processes. As a result, a
resource query initializes a learning resource recommendation process as shown in
Figure 7.
We propose a collaborative resource recommendation algorithm based on
k-nearest neighbour with the following stages as shown in Figure 8:
Table 5. ā€˜ā€˜Opinion-ratingā€™ā€™ policy for grading the resources.
Intensity, d g(d) l g(l) Opinion keywords, o g(o)
ā€˜ā€˜Slightlyā€™ā€™ 2 ā€˜ā€˜ā€“ā€™ā€™ 1 Negative ā€˜ā€˜boringā€™ā€™ 71
ā€˜ā€˜Moderatelyā€™ā€™ 3 ā€˜ā€˜toughā€™ā€™
ā€˜ā€˜Veryā€™ā€™ 4 ā€˜ā€˜notā€™ā€™ 71 ā€˜ā€˜badā€™ā€™
ā€˜ā€˜Extremelyā€™ā€™ 5 ā€˜ā€˜sillyā€™ā€™
ā€˜ā€˜lousyā€™ā€™
Ambivalent ā€˜ā€˜no commentā€™ā€™ 0
Positive ā€˜ā€˜coolā€™ā€™
ā€˜ā€˜interestingā€™ā€™
ā€˜ā€˜goodā€™ā€™ 1
ā€˜ā€˜usefulā€™ā€™
ā€˜ā€˜funā€™ā€™
S.B.-Y. Lau et al.
92
(1) Stage 1 ā€“ resource matching: the purpose of resource matching is to compute
semantic relevance of learning resources based on a learnerā€™s context proļ¬le
and pedagogical need speciļ¬ed in a resource query. Resources whose tags
fulļ¬l the query of the current active learner are shortlisted. The degree of
matching of resources is carried out by computing semantic relevance
measures between the learnerā€™s need (preferred theme, content format and
complexity level) and features of learning resources as tagged by users based
on framework of the semantic model for tags presented in Section 7. The level
of matching of a resource follows the function:
Smatching Ā¼ a  sTheme Ć¾ b  sFormat Ć¾ Ć°1  a  bƞ  sComplexity Ć°3ƞ
where Smatching 2 [ 0,1], coeļ¬ƒcient for Theme, b is the coeļ¬ƒcient for Content
Format, sTheme is the similarity between themes, sFormat is the similarity
between content formats and sComplexity is the similarity between the
complexity levels. Resources with Smatching 4 mth, a matching cut-oļ¬€ thresh-
old value will be considered to be relevant to a learnerā€™s query shortlisted for
next stage of processing.
(2) Stage 2 ā€“ resource classiļ¬cation: resources are classiļ¬ed into binary classes:
{ā€˜ā€˜Goodā€™ā€™, ā€˜ā€˜Poorā€™ā€™} based on the aggregated similarity measures of context
proļ¬les of the current active learner and the learners who tag the resources as
well as the average rating inferred from tags. Resources are classiļ¬ed as ā€˜ā€˜Goodā€™ā€™
if the predicted composite rating scores are above a ā€˜ā€˜Goodā€™ā€™ness threshold.
Deļ¬nition 1 (ā€˜ā€˜Goodā€™ā€™ or ā€˜ā€˜Poorā€™ā€™ Resources): Resources fulļ¬l the basic criteria
of being relevant to the theme, content format and interest of the learner, as
well as with high quality content. Technically, ā€˜ā€˜Goodā€™ā€™ness, grtij means the
computed rating score is greater than a ā€˜ā€˜Goodā€™ā€™ness threshold value, i.e.
grtij 4 gth. The ā€˜ā€˜Goodā€™ā€™ness value is an aggregation of relevance and quality
rating of content. Hence, if grtij  gth, a resource will be classiļ¬ed as ā€˜ā€˜Poorā€™ā€™.
Figure 7. Overview of the personalized learning process.
Interactive Learning Environments 93
Deļ¬nition 2 (Resource Classiļ¬cation): A resource from a sample set of
resources, R Ā¼ {r1, r2, . . . , rk}, is mapped to two classes C Ā¼ {c1, c2} Ā¼
{ā€˜ā€˜Goodā€™ā€™, ā€˜ā€˜Poorā€™ā€™} by a classiļ¬cation function f: ri ! cj 2 C. The classiļ¬ca-
tion function is weighted on the similarities between learnersā€™ context proļ¬les:
(learning style, competency level). Resources aggregately rated higher than
the threshold value, gth by similar learners and other dissimilar learners will
be classiļ¬ed as ā€˜ā€˜Goodā€™ā€™, while others are classiļ¬ed as ā€˜ā€˜Poorā€™ā€™.
Classiļ¬cation based on a rating function takes into account the similarity of
the taggerā€™s context proļ¬les: (learning style, competency level) to the current
active learner and the aggregate ratings of the resources weighted by learnersā€™
similarities:
Aggregated rating of a resource,
grtj Ā¼
1
N
X
N
iĀ¼1
wi  gi ư4ƞ
Figure 8. Overview of the resource recommendation algorithm.
S.B.-Y. Lau et al.
94
where N is the total number of rating instances, gi is the rating by ith learner
on the resource j, and wi is the weight factor for ri based on learner
similarities. Similarity or distance measure between ith learner and the current
active learner can be the value for wi to distinguish between the relative
signiļ¬cance of ratings imposed by learners. We may also use two distinct
values for w where a much larger w, e.g. 100, is used for similar learners
(nearest neighbours) compared to smaller w, e.g. 10, for dissimilar learners.
(3) Stage 3 ā€“ resource ļ¬ltering: if the rating score computed for a learning
resource as grti 4 gth, a resource is considered to be a ā€˜ā€˜Goodā€™ā€™ resource. Else,
it is excluded from further consideration as shown in Figure 9. A set of
ā€˜ā€˜Goodā€™ā€™ resources, R0
are considered for ranking.
(4) Stage 4 ā€“ resource ranking: resources, which are classiļ¬ed as ā€˜ā€˜Goodā€™ā€™ are
ranked in the descending order of predicted rating or degree of ā€˜ā€˜goodnessā€™ā€™.
(5) Stage 5 ā€“ resource presentation: ranked list of resources are presented to the
current active learner for interaction.
8. User study
To verify the eļ¬€ectiveness of the semantic model for tags, a prototype Web-based
learning resource recommender system has been implemented with PHP-MySQL
which is accessible via URLs such as http://3ants.vacau.com. Architecture of the
prototype learning resource recommender system is shown in Figure 10. The
Figure 9. Resource classiļ¬cation and ļ¬ltering.
Interactive Learning Environments 95
prototype allows users to share and tag hypermedia resources obtained from the
Internet as shown in Figure 11.
Users are able to annotate resources with concept(s) or theme(s) for a knowledge
domain such as ā€˜ā€˜personal ļ¬nanceā€™ā€™, the content type of the resource and the
complexity of the resources. Besides, the key novelty of the prototype system is the
subjective opinions expressed by users that can be automatically converted to
numerical ratings via the use of commonly found vocabularies of emotion as shown
in Figure 12 following an opinion-rating policy described in Section 6.4.
The purpose of this pilot user study is aimed at quantifying the various aspects of
quality of the recommendation results and the recommender system from the
perspectives of the users. As far as we know, currently there is yet any comprehensive
framework to evaluate a recommender system for the education domain. Hence, we
formulate a novel seven-pillar evaluation framework for learning recommender
system as shown in Figure 13. The evaluation framework is formulated via a
rigorous process as shown in Figure 14.
The system is assessed via questionnaires that encompass evaluation metrics such
as user-perceived accuracy, novelty and serendipity, usefulness, domain coverage and
diversity of information, conļ¬dence and trust of user, and information adequacy and
usability of a learning recommender system. Each of the metrics is explained brieļ¬‚y
as follows:
(1) User-perceived accuracy is a measure of the level of relevance of the
recommender results to what is being expected by users. Though accuracy
metrics of recommender systems such as precision, recall and F-measure are
more readily measurable via simulated data, it is helpful to conļ¬rm the
relevance of the recommended results judged by users.
(2) Novelty and serendipity of a recommender system ensure that users have
access to non-obvious, unexpected and ā€˜ā€˜surprisingly interestingā€™ā€™ items.
Figure 10. Resource sharing and tagging.
S.B.-Y. Lau et al.
96
(3) Domain coverage or information diversity is a measure of the size of the
domain of resources over which the system can make recommendations
which may include the content format, themes and topics towards fulļ¬lling
the learning needs.
(4) It is assumed that higher conļ¬dence and trust in the recommendation would
lead to more recommendations being used.
(5) Information adequacy is concerned with why an item is recommended as
adequately explained and conveyed to the users.
(6) Usefulness or eļ¬€ectiveness of the recommendation is concerned with whether
the recommendation is suitable to the information needs, learning goal or
task performed by the users.
(7) Usability is concerned with whether users enjoy the features of the user
interface of the recommender system and whether users perceive the tasks as
easy to complete. It is more a measure for the recommender system rather
than the recommendation results.
The pilot user study was carried out as a semi-controlled ļ¬eld study with users in
a social Web environment. The user study was piloted for the topic: ā€˜ā€˜Introduction to
personal ļ¬nanceā€™ā€™. Invitation to pilot test the prototype system was broadcasted to
over 100 direct or indirect friends on the social network. Friends who accepted the
Figure 11. Annotating with tags expressing subjective feelings about resource.
Interactive Learning Environments 97
Figure 12. Architecture of the learning resource recommender system.
Figure 13. The seven-pillar evaluation framework.
S.B.-Y. Lau et al.
98
invitation would log in to the prototype system for pilot runs at a time and place
convenient to them over a period of 2 weeks. At the end of the period of experiment,
there are a total of 186 resources shared and tagged by a total of 92 friends who
accessed the system. However, only about less than half (or 42%) of the respondents
completed the learning style and competency-level assessments. The age group of all
the respondents is between 20 and 40 with more than half completed or currently
pursuing tertiary education.
At the end of the process as shown in Figure 15, users went through a 30-
question evaluation based on the seven-pillar evaluation framework for the
recommended results and recommender system. For reasons unknown, only 11 of
the respondents completed the evaluation cycle. The evaluation results are
summarized in Table 6.
Although the sample size may not be statistically signiļ¬cant at this stage (where
only 11 respondents completed the evaluation cycle), the user evaluation results
provide useful hints into the factors that aļ¬€ect usersā€™ perception on the quality of
recommendation. Preliminary results show that the recommender system and the
recommendation results are rated in the positive territory (more than 3.5 from a scale
of 1ā€“7) overall, including the relevance of the recommended results (accuracy). The
recommendation generally gains conļ¬dence and trust from the users by scoring more
than 5 out of 7. Based on Table 6, the recommender is rated between 4.6 and 5 for
other quality criteria such as novelty, serendipity, domain coverage and information
diversity. As far as we know, as of today, there is no any traditional tagging system
speciļ¬cally designed for the educational domain. Eļ¬€ort is underway to build a
learning resource recommender system with the traditional tagging approach, and
more interesting results will be presented in upcoming publications.
9. Conclusion
To sum up, this article has successfully demonstrated the formulation of a
lightweight folksonomy-based resource annotation metadata schema based on a
semantic model for tags. The framework includes not only a domain that is formally
consistent, but is also accurate, extensible and current (based on expertsā€™
assessment). Most importantly, the testing of this framework provides the
underlying guiding principles for collaborative annotation of learning resources
using semantic tags. The semantic model and metadata schema form essential parts
of a personalized learning resource recommendation. The semantic model and
Figure 14. Formulation steps of the seven-pillar evaluation framework.
Interactive Learning Environments 99
Figure 15. Overview of the procedure for the execution of user study.
S.B.-Y. Lau et al.
100
Table 6. Evaluation results for user study.
No. Statement
Average
rating
A Accuracy
1 The items recommended to me matched my needs/preferences. 4.7692
2 The recommended items I received better-ļ¬ts my needs/preferences
than what I search using a search engine.
4.8462
3 The recommended results are ranked in the correct order of
relatedness to my needs.
4.6923
4 The recommended results are ranked in the correct order of rated
quality (in terms of usefulness of the item).
4.6154
Average 4.7308
B Novelty and serendipity
5 The system suggested unexpected new items that are relevant to my
needs and preferences.
4.6923
6 The recommender gave me good suggestions. 4.6923
Average 4.6923
C Domain coverage and information diversity
7 The items recommended to me cover diverse content types. 4.2308
8 The items recommended to me cover diverse themes/topics. 4.7692
9 There is enough diversity of items in the results to fulļ¬l my learning
needs.
4.6154
Average 4.5385
D Conļ¬dence and trust
10 I trust the recommended items that are ranked correctly. 5.0769
11 I am conļ¬dent in the quality of the items recommended to me. 5.0000
Average 5.0385
E Information adequacy
12 The information provided for the recommended items is suļ¬ƒcient for
me.
4.8461
13 I understood the reason why the items were recommended to me. 4.6923
Average 4.7692
F Usefulness and eļ¬€ectiveness
14 The recommender helped me be more eļ¬€ective in learning. 5.0000
15 The recommender helped me be more productive in learning. 5.0000
16 The recommended items inļ¬‚uenced my way of learning. 4.7692
17 I gained new knowledge/skills interacting with the recommended
items.
5.0769
18 I am positively stimulated/encouraged by the system to pursue the
learning of the subject matter further.
5.0000
19 The system was fun. 4.9231
20 The recommender helped me to decide better. 4.7692
Average 4.9341
G Usability
21 It was easy to look for and access a recommended item(s) I need/like. 4.9231
22 The system was ļ¬‚exible. The recommender provides an adequate way
for me to express and revise my needs/preferences.
4.9231
23 I felt comfortable using this system. 4.8461
24 It was easy for me to inform the system if I disliked/liked the
recommended item (provide feedback).
4.7692
Average 4.8654
(continued)
Interactive Learning Environments 101
metadata schema provide the framework for item-based ļ¬ltering of learning
resources prior to a user-based classiļ¬cation of resources as well as ranking of
learning resources based on numerical ratings inferred from tags within a
collaborative resource recommendation algorithm based on the k-nearest neighbour.
At present, as a proof-of-concept, the semantic model and metadata schema are
implemented in a prototype Web-based learning resource recommender system.
Preliminary results of an ongoing user study show that associating collaborative
tagging with controlled vocabularies in learning resource annotation is eļ¬€ective in
personalizing learning resources. Under the current framework, a learner is able to
leverage the opinions of others in order to locate the right, relevant resource of good
quality while avoiding less useful ones.
Notes on contributors
Simon Boung-Yew Lau is a lecturer and researcher at the Faculty of Engineering and Science,
Universiti Tunku Abdul Rahman, Kuala Lumpur, Malaysia. He earned his master of science
degree from the Malaysia University of Science and Technology. He teaches undergraduate
and postgraduate courses, specializing in software engineering. His research areas of interest
include context-aware computing, e-learning, social Web and parallel computing.
Chien-Sing Leeā€™s research spans across instructional design, computer-supported collaborative
learning, game-based learning, mobile learning, data mining and intelligent agents. She has
experimented with the use of service-oriented architectures to deliver personalized instruction
and the use of knowledge management not only to organize knowledge but also to derive best
practices from prior knowledge so that these best practices can be easily adapted to similar
scenarios. Another area of interest is interoperability through the semantic Web due to her
concern that best practices should be shared, adopted and adapted locally and globally to
improve teaching and learning practices in half the time half the cost but with the highest
eļ¬€ectiveness and eļ¬ƒciency. Her latest experiments deal with scaļ¬€olding creativity in learning
environments.
Yashwant Prasad Singh received the B.Sc.Engg. (Electrical) degree (ļ¬rst-class honours) from
Bihar College of Engineering, Bhagalpur, India, the M.E. degree in electronics and
communications (ļ¬rst-class honours) from the University of Roorkee, Roorkee, India, and
the Ph.D. degree in electrical engineering from the Indian Institute of Technology, Kanpur,
India, in 1966, 1975 and 1980, respectively. He is currently a professor of computer
engineering, Faculty of Information Technology, Multimedia University, Cyberjaya,
Table 6. (Continued)
No. Statement
Average
rating
H Overall
25 Overall, I am satisļ¬ed with the recommender. 5.0000
26 I am always conļ¬dent I will like the items recommended to me. 4.7692
27 The system was able to convince me about the goodness of the
recommendations.
5.0000
Average 4.9231
I Continuance and frequency (future use)
28 I will use this recommender again. 4.7692
29 I will use this recommender frequently. 4.7692
Average 4.7692
J Recommendation to friends
30 I will recommend this recommender to my friends. 5.0000
Overall average 4.8261
S.B.-Y. Lau et al.
102
Malaysia. His research interests include complexity theory, cognitive science, computational
neuroscience, Artiļ¬cial Neural Network (ANN), fuzzy systems, artiļ¬cial intelligence algorithms
and intelligent systems, machine learning and data mining, natural computation (evolutionary
computing, neural computation and artiļ¬cial immune systems) and their applications to
classiļ¬cation, clustering, anomaly detection, association rule mining, ranking and routing, as well
as computer and microprocessor architectures, including transputers. He has published more
than 100 papers in international and national journals and conference proceedings.
Acknowledgement
This work is part of the ļ¬rst authorā€™s PhD research conducted partly in Multimedia University
and partly while the second author was a faculty at Multimedia University.
References
Altfest, L.J. (2007). Personal ļ¬nancial planning. Boston, MA: McGraw-Hill Irwin.
Bajtelsmit, V.L. (2006). Personal ļ¬nance: Skills for life. Hoboken, NJ: John Wiley.
Blecksmith, R. (2010). Personal ļ¬nance lecture notes continued. Retrieved from http://www.
math.niu.edu/*richard/Math101/sp07/monthly_savings_ho.pdf
Bouillet, E., Feblowitz, M., Liu, Z., Ranganathan, A.,  Riabov, A. (2008). A tag-based
approach for the design and composition of information processing applications. In
Proceedings of the 23rd ACM SIGPLAN conference on object-oriented programming
systems languages and applications (pp. 585ā€“602). doi: 10.1145/1449955.1449810
Buriano, L., Marchetti, M., Carmagnola, F., Cena, F., Gena, C.,  Torre, I. (2006). The role
of ontologies in context-aware recommender systems. In The 7th international conference
on mobile data management. doi: 10.1109/MDM.2006.149
Cantador, I., BellogıĢn, A.,  Castells, P. (2008). A multilayer ontology-based hybrid
recommendation model. AI Communications, 21(2ā€“3), 203ā€“210. doi: 10.1.1.144.4119.
Costa, A., Guizzardi, R., Guizzardi, G.,  Filho, J. (2007, June). COReS: Context-aware,
ontology-based recommender system for service recommendation. Paper presented at the
meeting of the 19th International Conference on Advanced Information Systems
Engineering, Trondheim, Norway.
DMOZ Open Directory Project. (2012). Retrieved from http://www.dmoz.org/
Downey, T. (2005). The Standard  Poorā€™s guide to personal ļ¬nance. New York, NY:
McGraw-Hill.
Drachsler, H., Hummel, H.G.K.,  Koper, R. (2007)Recommendations for learners are
diļ¬€erent: Applying memory-based recommender system techniques to lifelong learning. In
The SIRTEL workshop at the EC-TEL 2007 conference (pp. 17ā€“20).
Drachsler, H.,  Manouselis, N. (2009). How recommender systems in technology-enhanced
learning depend on context. In The 1st Workshop on Context-aware Recommender Systems
for Learning at the Alpine Rendez-Vous 2009.
Fok, A.W.P.,  Ip, H.H.S. (2007). Educational ontologies construction for personalized
learning on the web. Studies in Computational Intelligence, 82, 47ā€“82. doi: 10.1007/978-3-
540-71974-8_4
Frasca, R.R. (2009). Personal ļ¬nance: An integrated planning approach. Boston, MA: Prentice
Hall.
Gadepalli, G., Rundensteiner, E., Brown, D.,  Claypool, K. (2010). Tag and resource-aware
collaborative ļ¬ltering algorithms for resource recommendation. In The 2010 seventh
international conference on information technology (pp. 546ā€“551). doi: 10.1109/ITNG.
2010.48
Grafman Productions. (1997). Internet media type. Retrieved from http://graphcomp.com/
info/specs/mime.html
Guy, M.,  Tonkin, E. (2006). Folksonomies: Tidying up tags? Retrieved from http://www.
dlib.org/dlib/january06/guy/01guy.html
Heiyanthuduwage, S.R.,  Karunaratne, D.D. (2006). A learner oriented ontology of
metadata to improve eļ¬€ectiveness of learning management systems. In The 3rd inter-
national conference on eLearning for knowledge-based society, Bangkok, Thailand. doi:
10.1.1.136.8840
Interactive Learning Environments 103
Itmazi, J.,  Gea, M. (2006). The recommendation systems: Types, domains and the ability
usage in learning management system. In The international Arab conference on information
technology (ACITā€™2006).
Itmazi, J.,  Megias, M. (2008). Using recommender system in course management systems to
recommend learning objects. The International Arab Journal of Information Technology,
5(3), 234ā€“240.
Jovanovic, J., GasĢŒevicĢ, D.,  DevedzĢŒicĢ, V. (2009). TANGRAM for personalized learning
using the semantic web technologies. Journal of Emerging Technologies in Web Intelligence,
1(1), 6ā€“21. doi: 10.4304/jetwi.1.1.6-21
Kapoor, J.R. (2009). Personal ļ¬nance. Boston, MA: McGraw-Hill Irwin.
Karampiperis, P.,  Sampson, D. (2005). Adaptive learning resources sequencing in educa-
tional hypermedia systems. Educational Technology  Society. 8, 128ā€“147. doi: 10.1.
1.98.5688
Keown, J. (2010). Financial planning. Retrieved from http://www.ļ¬nance.pamplin.vt.edu/
faculty/ajk/pfm/html/lecture.htm
Kerkiri, T., Manitsaris, A.,  Mavridou, A. (2007). Reputation metadata for recommending
personalized e-learning resources. In The second international workshop on semantic media
adaptation and personalization (pp. 110ā€“115). doi: 10.1109/SMAP.2007.38
Khribi, M.K., Jemni, M.,  Nasraoui, O. (2007). Toward a hybrid recommender system for e-
learning personalization based on web usage mining techniques and information retrieval.
In 2007 world conference on e-learning in corporate, government, healthcare, and higher
education (pp. 6136ā€“6145).
Khribi, M.K., Jemni, M.,  Nasraoui, O. (2009). Automatic recommendations for e-learning
personalization based on web usage mining techniques and information retrieval.
Educational Technology  Society, 12(4), 30ā€“42. doi: 10.1109/ICALT.2008.198
Kim, H.-L., Breslin, J.G., Yang, S.-K.,  Kim, H.-G. (2008). Social semantic cloud of tag:
Semantic model for social tagging. In The 2nd KES international symposium on agent and
multi-agent systems: Technologies and applications, Incheon, Korea (pp. 83ā€“92). doi:
10.1007/978-3-540-78582-8_9
Lau, B.-Y.-S.,  Lee, C.S. (2012). Enhancing collaborative ļ¬ltering of learning resources with
semantically-enhanced social tags. In The 12th IEEE international conference on advanced
learning technologies (ICALT2012) (pp. 281ā€“285). Washington, DC: IEEE.
Leo, O., Ceusters, W., Mani, I., Ray S.,  Smith, B. (2007). The evaluation of ontologies:
Toward improved semantic interoperability. Semantic Web: Revolutionizing knowledge
discovery in the Life Sciences, 139ā€“158. doi: 10.1.1.82.1024
Li, Q.-F.,  Lu, S.C.-Y. (2008). Collaborative tagging applications and approaches. IEEE
Multimedia, 15(3), 14ā€“21. doi: 10.1109/MMUL.2008.54
Lin, H.-N., Tseng, S.-S., Weng, J.-F., Lin, H.-Y.,  Su, J.-M. (2007). An iterative, collaborative
ontology construction scheme. In The 2nd international conference on innovative computing,
information and control (ICICIC 2007) (pp. 150). doi: 10.1109/ICICIC.2007.164
Marchetti, A., Tesconi, M.,  Ronzano, F. (2007). SemKey: A semantic collaborative tagging
system. Paper presented at the meeting of the Workshop, Tagging and Metadata for Social
Information Organization, Banļ¬€, Alberta, Canada.
Mathes, A. (2004). Folksonomies ā€“ Cooperative classiļ¬cation and communication through shared
metadata. Retrieved from http://www.adammathes.com/academic/computer-mediated-
communication/folksonomies.html
Michael, J. (1997). Personal ļ¬nance on the web: An interactive guide. New York, NY: Wiley.
Mika, P. (2005). Ontologies are us: A uniļ¬ed model of social networks and semantics. In
International semantic web conference (pp. 522ā€“536). doi: 10.1016/j.websem.2006.11.002
Milicevic, A., Nanopoulos, A.,  Ivanovic, M. (2010). Social tagging in recommender systems:
A survey of the state-of-the-art and possible extensions. Artiļ¬cial Intelligence Review,
33(3), 187ā€“209. doi: 10.1007/s10462-009-9153-2
Nakamoto, R., Nakajima, S., Miyazaki, J.,  Uemura, S. (2008). Tag-based context-
ual collaborative ļ¬ltering. IAENG International Journal of Computer Science, 34(2), 214ā€“219.
Naudet, Y., Mignon, S., Lecaque, L., Hazotte, C.,  Groues, V. (2008). Ontology-based
matchmaking approach for context-aware recommendations. In International conference
on automated solutions for cross media content and multi-channel distribution (pp. 218ā€“223).
IEEE Xplorer Digital Library. doi: 10.1109/AXMEDIS.2008.13
S.B.-Y. Lau et al.
104
Otair, M.A.,  Al Hamad, A.Q. (2005). Expert personalized e-learning recommender system.
In The 2005 international conference on e-business and e-learning (EBEL 2005).
Pfeiļ¬€er, H.,  Tonkin, E. (2009). Tagging in context: Information management across
community networks. In C. Bouras, V. Poulopoulos,  V. Tsogkas (Eds.), Handbook of
research on social interaction technologies and collaboration software: Concepts and trends.
USA: IGI Global. doi: 10.4018/978-1-60566-368-5.ch015
Pinto, H.S.,  Peralta, D.N. (2003). Combining ontology engineering subprocesses to build a
time ontology. In The 2nd International Conference On Knowledge Capture (pp. 88ā€“95).
doi: 10.1145/945645.945660
Quintarelli, E. (2005). Folksonomies: Power to the people. Retrieved from http://www.iskoi.
org/doc/folksonomies.htm
Santos, O.C., Baldiris, S., Boticario1, J.G., Restrepo1, E.G.,  Fabregat, R. (2011). Open
issues in personalized inclusive learning scenarios. In The international workshop on
personalization approaches in learning environment (pp. 54ā€“58).
Santos O.C., , Boticario J.G. (2011). Requirements for semantic educational recommender
systems in formal e-learning scenarios. Algorithms, 4(2), 131ā€“154. doi: 10.3390/a4030131
Shih, B.-Y.,  Lee, W.-I. (2001). The application of nearest neighbour algorithm on creating
an adaptive on-line learning system. In The 31th ASEE/IEEE frontiers in education
conference (pp. 10ā€“13). doi: http://doi.ieeecomputersociety.org/10.1109/FIE.2001.963925
Shroļ¬€, M.P. (2011, July 17). Educating the facebook generation. Daily news  analysis. Retrieved
from http://www.dnaindia.com/india/column_educating-the-facebook-generation_1566688
SĢŒimuĢn, M., Andrejko, A.,  BielikovaĢ, M. (2007). Ontology-based models for personalized e-
learning environment. Paper presented at the meeting of the 5th International Conference
on Emerging e-Learning Technologies and Applications.
Soonthornphisaj, N., Rojsattarat, E.,  Yim-ngam, S. (2006). Smart e-learning using recom-
mender system. In Computational Intelligence (pp. 518ā€“523). Berlin: Springer.
Swarup, S.,  Gasser, L. (2008). Collaborative tagging as information foraging (Technical
Report UIUCLISā€“2008/1Ć¾ID). Urbana-Champaign: University of Illinois, Graduate
School of Library and Information Science.
Tan, H.-Y., Guo, J.-F.,  Li, Y. (2008). E-learning recommendation system. In International
conference on computer science and software engineering (pp. 430ā€“433). doi: http://doi.ieee
computersociety.org/10.1109/CSSE.2008.305
Tang, T.,  McCalla, G. (2005). Smart recommendation for an evolving e-learning system:
Architecture and experiment. International Journal on E-Learning, 4(1), 105ā€“129.
Tsai, K.-H., Chiu, T.-K., Lee, M.-C.,  Wang, T.I. (2006). A learning objects recommenda-
tion model based on the preference and ontological approaches. In The sixth international
conference on advanced learning technologies. doi: http://doi.ieeecomputersociety.org/
10.1109/ICALT.2006.18
Tso-Sutter, K.H.L., Marinho, L.B.,  Schmidt-Thieme, L. (2008). Tag-aware recommender
systems by fusion of collaborative ļ¬ltering algorithms. In The 23rd annual ACM
symposium on applied computing (pp. 1995ā€“1999). doi: 10.1145/1363686.1364171
Ulbrich, A., Kandpal, D.,  Tochtermann, K. (2003). First steps towards personalization
concepts in e-learning. In Wissensmanagement (pp. 229ā€“233).
Uschold, M.,  GruĢˆninger, M. (1996). Ontologies: Principles methods and applications.
Knowledge Sharing and Review, 2, 93ā€“136.
Vanderwal, T. (2007). Folksonomy: Folksonomy coinage and deļ¬nition. Retrieved from http://
vanderwal.net/folksonomy.html
Xu,Z.,Fu,Y.,Mao,J.,Su,D.(2006).Towardsthesemanticweb:Collaborativetagsuggestions.Paper
presented at the meeting of the Collaborative Web Tagging Workshop, Edinburgh, Scotland.
Yun, H.-Y. (2009). Research on building ocean domain ontology. In The Workshop on
Computer Science and Engineering, Vol. 1 (pp. 146ā€“150). Washington, DC: IEEE
Computer Society. doi: 10.1109/WCSE.2009.641
Yun, H.-Y., Xu, J.-L., Wei, M.-J.,  Xiong, J. (2009). Development of domain ontology for e-
learning course. In 2009 IEEE international symposium on IT in medicine and education
(ITME2009) (pp. 501ā€“506). doi: 10.1109/ITIME.2009.5236370
Zhang, Z.K.,  Chuang, L. (2010). A hypergraph model of social tagging networks. Journal of
Statistical Mechanics: Theory and Experiment, 10, 2ā€“14. doi: 10.1088/1742ā€“5468/2010/10/
P10005
Interactive Learning Environments 105
Copyright of Interactive Learning Environments is the property of Routledge and its content
may not be copied or emailed to multiple sites or posted to a listserv without the copyright
holder's express written permission. However, users may print, download, or email articles for
individual use.

More Related Content

Similar to Folksonomy-based lightweight resource annotation metadata schema for personalized hypermedia learning

Running Head ANNOTATED BIBLIOGRAPHY1ANNOTATED BIBLIOGRAPHY .docx
Running Head ANNOTATED BIBLIOGRAPHY1ANNOTATED BIBLIOGRAPHY .docxRunning Head ANNOTATED BIBLIOGRAPHY1ANNOTATED BIBLIOGRAPHY .docx
Running Head ANNOTATED BIBLIOGRAPHY1ANNOTATED BIBLIOGRAPHY .docxSUBHI7
Ā 
Enriching E-Learning with web Services for the Creation of Virtual Learning P...
Enriching E-Learning with web Services for the Creation of Virtual Learning P...Enriching E-Learning with web Services for the Creation of Virtual Learning P...
Enriching E-Learning with web Services for the Creation of Virtual Learning P...IJERDJOURNAL
Ā 
Designing a Scaffolding for Supporting Personalized Synchronous e-Learning
Designing a Scaffolding for Supporting Personalized Synchronous e-LearningDesigning a Scaffolding for Supporting Personalized Synchronous e-Learning
Designing a Scaffolding for Supporting Personalized Synchronous e-Learningcscpconf
Ā 
EMBODYING THE EDUCATIONAL INSTITUTION
EMBODYING THE EDUCATIONAL INSTITUTIONEMBODYING THE EDUCATIONAL INSTITUTION
EMBODYING THE EDUCATIONAL INSTITUTIONijejournal
Ā 
Final conole southampton
Final conole southamptonFinal conole southampton
Final conole southamptongrainne
Ā 
Context modelling for Learning; some heuristics
Context modelling for Learning; some heuristicsContext modelling for Learning; some heuristics
Context modelling for Learning; some heuristicsLondon Knowledge Lab
Ā 
A design of a multi-agent recommendation system using ontologies and rule-bas...
A design of a multi-agent recommendation system using ontologies and rule-bas...A design of a multi-agent recommendation system using ontologies and rule-bas...
A design of a multi-agent recommendation system using ontologies and rule-bas...IJECEIAES
Ā 
FUNCTIONAL SEMANTICS AWARE BROKER BASED ARCHITECTURE FOR E-LEARNING WEB SERVICES
FUNCTIONAL SEMANTICS AWARE BROKER BASED ARCHITECTURE FOR E-LEARNING WEB SERVICESFUNCTIONAL SEMANTICS AWARE BROKER BASED ARCHITECTURE FOR E-LEARNING WEB SERVICES
FUNCTIONAL SEMANTICS AWARE BROKER BASED ARCHITECTURE FOR E-LEARNING WEB SERVICESIJITE
Ā 
Social web and language learning
Social web and language learningSocial web and language learning
Social web and language learningEsperanza RomƔn
Ā 
Semantically Enchanced Personalised Adaptive E-Learning for General and Dysle...
Semantically Enchanced Personalised Adaptive E-Learning for General and Dysle...Semantically Enchanced Personalised Adaptive E-Learning for General and Dysle...
Semantically Enchanced Personalised Adaptive E-Learning for General and Dysle...Eswar Publications
Ā 
Solving The Problem of Adaptive E-Learning By Using Social Networks
Solving The Problem of Adaptive E-Learning By Using Social NetworksSolving The Problem of Adaptive E-Learning By Using Social Networks
Solving The Problem of Adaptive E-Learning By Using Social NetworksEswar Publications
Ā 
Language Translation for E-learning Systems
Language Translation for E-learning SystemsLanguage Translation for E-learning Systems
Language Translation for E-learning SystemsIRJET Journal
Ā 
Collaborative Learning of Organisational Knolwedge
Collaborative Learning of Organisational KnolwedgeCollaborative Learning of Organisational Knolwedge
Collaborative Learning of Organisational KnolwedgeWaqas Tariq
Ā 
IJCER (www.ijceronline.com) International Journal of computational Engineerin...
IJCER (www.ijceronline.com) International Journal of computational Engineerin...IJCER (www.ijceronline.com) International Journal of computational Engineerin...
IJCER (www.ijceronline.com) International Journal of computational Engineerin...ijceronline
Ā 
OLAP based Scaffolding to support Personalized Synchronous e-Learning
 OLAP based Scaffolding to support Personalized Synchronous e-Learning  OLAP based Scaffolding to support Personalized Synchronous e-Learning
OLAP based Scaffolding to support Personalized Synchronous e-Learning IJMIT JOURNAL
Ā 
Qi bl 2014 wienerneustadt quantitative and qualitative criteria 0.8
Qi bl 2014 wienerneustadt quantitative and qualitative criteria 0.8Qi bl 2014 wienerneustadt quantitative and qualitative criteria 0.8
Qi bl 2014 wienerneustadt quantitative and qualitative criteria 0.8Stefano Lariccia
Ā 
Miller - A System for Integrating Online Multimedia into College Curriculum ...
Miller -  A System for Integrating Online Multimedia into College Curriculum ...Miller -  A System for Integrating Online Multimedia into College Curriculum ...
Miller - A System for Integrating Online Multimedia into College Curriculum ...ut san antonio
Ā 

Similar to Folksonomy-based lightweight resource annotation metadata schema for personalized hypermedia learning (20)

Running Head ANNOTATED BIBLIOGRAPHY1ANNOTATED BIBLIOGRAPHY .docx
Running Head ANNOTATED BIBLIOGRAPHY1ANNOTATED BIBLIOGRAPHY .docxRunning Head ANNOTATED BIBLIOGRAPHY1ANNOTATED BIBLIOGRAPHY .docx
Running Head ANNOTATED BIBLIOGRAPHY1ANNOTATED BIBLIOGRAPHY .docx
Ā 
Enriching E-Learning with web Services for the Creation of Virtual Learning P...
Enriching E-Learning with web Services for the Creation of Virtual Learning P...Enriching E-Learning with web Services for the Creation of Virtual Learning P...
Enriching E-Learning with web Services for the Creation of Virtual Learning P...
Ā 
Applying Semantic Web Technologies to Services of e-learning System
Applying Semantic Web Technologies to Services of e-learning SystemApplying Semantic Web Technologies to Services of e-learning System
Applying Semantic Web Technologies to Services of e-learning System
Ā 
D04 06 2438
D04 06 2438D04 06 2438
D04 06 2438
Ā 
Designing a Scaffolding for Supporting Personalized Synchronous e-Learning
Designing a Scaffolding for Supporting Personalized Synchronous e-LearningDesigning a Scaffolding for Supporting Personalized Synchronous e-Learning
Designing a Scaffolding for Supporting Personalized Synchronous e-Learning
Ā 
EMBODYING THE EDUCATIONAL INSTITUTION
EMBODYING THE EDUCATIONAL INSTITUTIONEMBODYING THE EDUCATIONAL INSTITUTION
EMBODYING THE EDUCATIONAL INSTITUTION
Ā 
Spinger3
Spinger3Spinger3
Spinger3
Ā 
Final conole southampton
Final conole southamptonFinal conole southampton
Final conole southampton
Ā 
Context modelling for Learning; some heuristics
Context modelling for Learning; some heuristicsContext modelling for Learning; some heuristics
Context modelling for Learning; some heuristics
Ā 
A design of a multi-agent recommendation system using ontologies and rule-bas...
A design of a multi-agent recommendation system using ontologies and rule-bas...A design of a multi-agent recommendation system using ontologies and rule-bas...
A design of a multi-agent recommendation system using ontologies and rule-bas...
Ā 
FUNCTIONAL SEMANTICS AWARE BROKER BASED ARCHITECTURE FOR E-LEARNING WEB SERVICES
FUNCTIONAL SEMANTICS AWARE BROKER BASED ARCHITECTURE FOR E-LEARNING WEB SERVICESFUNCTIONAL SEMANTICS AWARE BROKER BASED ARCHITECTURE FOR E-LEARNING WEB SERVICES
FUNCTIONAL SEMANTICS AWARE BROKER BASED ARCHITECTURE FOR E-LEARNING WEB SERVICES
Ā 
Social web and language learning
Social web and language learningSocial web and language learning
Social web and language learning
Ā 
Semantically Enchanced Personalised Adaptive E-Learning for General and Dysle...
Semantically Enchanced Personalised Adaptive E-Learning for General and Dysle...Semantically Enchanced Personalised Adaptive E-Learning for General and Dysle...
Semantically Enchanced Personalised Adaptive E-Learning for General and Dysle...
Ā 
Solving The Problem of Adaptive E-Learning By Using Social Networks
Solving The Problem of Adaptive E-Learning By Using Social NetworksSolving The Problem of Adaptive E-Learning By Using Social Networks
Solving The Problem of Adaptive E-Learning By Using Social Networks
Ā 
Language Translation for E-learning Systems
Language Translation for E-learning SystemsLanguage Translation for E-learning Systems
Language Translation for E-learning Systems
Ā 
Collaborative Learning of Organisational Knolwedge
Collaborative Learning of Organisational KnolwedgeCollaborative Learning of Organisational Knolwedge
Collaborative Learning of Organisational Knolwedge
Ā 
IJCER (www.ijceronline.com) International Journal of computational Engineerin...
IJCER (www.ijceronline.com) International Journal of computational Engineerin...IJCER (www.ijceronline.com) International Journal of computational Engineerin...
IJCER (www.ijceronline.com) International Journal of computational Engineerin...
Ā 
OLAP based Scaffolding to support Personalized Synchronous e-Learning
 OLAP based Scaffolding to support Personalized Synchronous e-Learning  OLAP based Scaffolding to support Personalized Synchronous e-Learning
OLAP based Scaffolding to support Personalized Synchronous e-Learning
Ā 
Qi bl 2014 wienerneustadt quantitative and qualitative criteria 0.8
Qi bl 2014 wienerneustadt quantitative and qualitative criteria 0.8Qi bl 2014 wienerneustadt quantitative and qualitative criteria 0.8
Qi bl 2014 wienerneustadt quantitative and qualitative criteria 0.8
Ā 
Miller - A System for Integrating Online Multimedia into College Curriculum ...
Miller -  A System for Integrating Online Multimedia into College Curriculum ...Miller -  A System for Integrating Online Multimedia into College Curriculum ...
Miller - A System for Integrating Online Multimedia into College Curriculum ...
Ā 

More from Kim Daniels

8 Pcs Vintage Lotus Letter Paper Stationery Writing P
8 Pcs Vintage Lotus Letter Paper Stationery Writing P8 Pcs Vintage Lotus Letter Paper Stationery Writing P
8 Pcs Vintage Lotus Letter Paper Stationery Writing PKim Daniels
Ā 
Essay Writing Words 100 Useful Words And Phrase
Essay Writing Words 100 Useful Words And PhraseEssay Writing Words 100 Useful Words And Phrase
Essay Writing Words 100 Useful Words And PhraseKim Daniels
Ā 
Descriptive Essay Example Of Expository Essays
Descriptive Essay Example Of Expository EssaysDescriptive Essay Example Of Expository Essays
Descriptive Essay Example Of Expository EssaysKim Daniels
Ā 
Free Printable Primary Lined Writing Paper - Discover T
Free Printable Primary Lined Writing Paper - Discover TFree Printable Primary Lined Writing Paper - Discover T
Free Printable Primary Lined Writing Paper - Discover TKim Daniels
Ā 
Example Of Argumentative Essay Conclusion Sitedoct.Org
Example Of Argumentative Essay Conclusion Sitedoct.OrgExample Of Argumentative Essay Conclusion Sitedoct.Org
Example Of Argumentative Essay Conclusion Sitedoct.OrgKim Daniels
Ā 
WRITE THE DATES Gener
WRITE THE DATES GenerWRITE THE DATES Gener
WRITE THE DATES GenerKim Daniels
Ā 
How To Start An Essay Introduction About A Book Printers Copy
How To Start An Essay Introduction About A Book Printers CopyHow To Start An Essay Introduction About A Book Printers Copy
How To Start An Essay Introduction About A Book Printers CopyKim Daniels
Ā 
Shocking How To Write A Transfer Essay Thatsnotus
Shocking How To Write A Transfer Essay ThatsnotusShocking How To Write A Transfer Essay Thatsnotus
Shocking How To Write A Transfer Essay ThatsnotusKim Daniels
Ā 
Owl Writing Paper Differentiated By Loving Le
Owl Writing Paper Differentiated By Loving LeOwl Writing Paper Differentiated By Loving Le
Owl Writing Paper Differentiated By Loving LeKim Daniels
Ā 
Free Printable Letter To Santa - Printable Templates
Free Printable Letter To Santa - Printable TemplatesFree Printable Letter To Santa - Printable Templates
Free Printable Letter To Santa - Printable TemplatesKim Daniels
Ā 
Heart Writing Practice Or Story Paper For Valentine
Heart Writing Practice Or Story Paper For ValentineHeart Writing Practice Or Story Paper For Valentine
Heart Writing Practice Or Story Paper For ValentineKim Daniels
Ā 
What I Learned In Computer Class Essay. I Learned For My Programming
What I Learned In Computer Class Essay. I Learned For My ProgrammingWhat I Learned In Computer Class Essay. I Learned For My Programming
What I Learned In Computer Class Essay. I Learned For My ProgrammingKim Daniels
Ā 
Citing A Website In An Essay Mla - MLA Citation Guide (8Th Edition ...
Citing A Website In An Essay Mla - MLA Citation Guide (8Th Edition ...Citing A Website In An Essay Mla - MLA Citation Guide (8Th Edition ...
Citing A Website In An Essay Mla - MLA Citation Guide (8Th Edition ...Kim Daniels
Ā 
Arguing About Literature A Guide And Reader - Arguin
Arguing About Literature A Guide And Reader - ArguinArguing About Literature A Guide And Reader - Arguin
Arguing About Literature A Guide And Reader - ArguinKim Daniels
Ā 
30 Nhs Letter Of Recommendation Templat
30 Nhs Letter Of Recommendation Templat30 Nhs Letter Of Recommendation Templat
30 Nhs Letter Of Recommendation TemplatKim Daniels
Ā 
Writing Personal Essays Examples - Short Essay S
Writing Personal Essays Examples - Short Essay SWriting Personal Essays Examples - Short Essay S
Writing Personal Essays Examples - Short Essay SKim Daniels
Ā 
Printable Primary Writing Paper PrintableTempl
Printable Primary Writing Paper  PrintableTemplPrintable Primary Writing Paper  PrintableTempl
Printable Primary Writing Paper PrintableTemplKim Daniels
Ā 
Mla Format Template With Cover Page HQ Printabl
Mla Format Template With Cover Page  HQ PrintablMla Format Template With Cover Page  HQ Printabl
Mla Format Template With Cover Page HQ PrintablKim Daniels
Ā 
Reaction Paper Introduction Sample. Reaction Paper In
Reaction Paper Introduction Sample. Reaction Paper InReaction Paper Introduction Sample. Reaction Paper In
Reaction Paper Introduction Sample. Reaction Paper InKim Daniels
Ā 
Persuasive Essay Essay Writing Help Online
Persuasive Essay Essay Writing Help OnlinePersuasive Essay Essay Writing Help Online
Persuasive Essay Essay Writing Help OnlineKim Daniels
Ā 

More from Kim Daniels (20)

8 Pcs Vintage Lotus Letter Paper Stationery Writing P
8 Pcs Vintage Lotus Letter Paper Stationery Writing P8 Pcs Vintage Lotus Letter Paper Stationery Writing P
8 Pcs Vintage Lotus Letter Paper Stationery Writing P
Ā 
Essay Writing Words 100 Useful Words And Phrase
Essay Writing Words 100 Useful Words And PhraseEssay Writing Words 100 Useful Words And Phrase
Essay Writing Words 100 Useful Words And Phrase
Ā 
Descriptive Essay Example Of Expository Essays
Descriptive Essay Example Of Expository EssaysDescriptive Essay Example Of Expository Essays
Descriptive Essay Example Of Expository Essays
Ā 
Free Printable Primary Lined Writing Paper - Discover T
Free Printable Primary Lined Writing Paper - Discover TFree Printable Primary Lined Writing Paper - Discover T
Free Printable Primary Lined Writing Paper - Discover T
Ā 
Example Of Argumentative Essay Conclusion Sitedoct.Org
Example Of Argumentative Essay Conclusion Sitedoct.OrgExample Of Argumentative Essay Conclusion Sitedoct.Org
Example Of Argumentative Essay Conclusion Sitedoct.Org
Ā 
WRITE THE DATES Gener
WRITE THE DATES GenerWRITE THE DATES Gener
WRITE THE DATES Gener
Ā 
How To Start An Essay Introduction About A Book Printers Copy
How To Start An Essay Introduction About A Book Printers CopyHow To Start An Essay Introduction About A Book Printers Copy
How To Start An Essay Introduction About A Book Printers Copy
Ā 
Shocking How To Write A Transfer Essay Thatsnotus
Shocking How To Write A Transfer Essay ThatsnotusShocking How To Write A Transfer Essay Thatsnotus
Shocking How To Write A Transfer Essay Thatsnotus
Ā 
Owl Writing Paper Differentiated By Loving Le
Owl Writing Paper Differentiated By Loving LeOwl Writing Paper Differentiated By Loving Le
Owl Writing Paper Differentiated By Loving Le
Ā 
Free Printable Letter To Santa - Printable Templates
Free Printable Letter To Santa - Printable TemplatesFree Printable Letter To Santa - Printable Templates
Free Printable Letter To Santa - Printable Templates
Ā 
Heart Writing Practice Or Story Paper For Valentine
Heart Writing Practice Or Story Paper For ValentineHeart Writing Practice Or Story Paper For Valentine
Heart Writing Practice Or Story Paper For Valentine
Ā 
What I Learned In Computer Class Essay. I Learned For My Programming
What I Learned In Computer Class Essay. I Learned For My ProgrammingWhat I Learned In Computer Class Essay. I Learned For My Programming
What I Learned In Computer Class Essay. I Learned For My Programming
Ā 
Citing A Website In An Essay Mla - MLA Citation Guide (8Th Edition ...
Citing A Website In An Essay Mla - MLA Citation Guide (8Th Edition ...Citing A Website In An Essay Mla - MLA Citation Guide (8Th Edition ...
Citing A Website In An Essay Mla - MLA Citation Guide (8Th Edition ...
Ā 
Arguing About Literature A Guide And Reader - Arguin
Arguing About Literature A Guide And Reader - ArguinArguing About Literature A Guide And Reader - Arguin
Arguing About Literature A Guide And Reader - Arguin
Ā 
30 Nhs Letter Of Recommendation Templat
30 Nhs Letter Of Recommendation Templat30 Nhs Letter Of Recommendation Templat
30 Nhs Letter Of Recommendation Templat
Ā 
Writing Personal Essays Examples - Short Essay S
Writing Personal Essays Examples - Short Essay SWriting Personal Essays Examples - Short Essay S
Writing Personal Essays Examples - Short Essay S
Ā 
Printable Primary Writing Paper PrintableTempl
Printable Primary Writing Paper  PrintableTemplPrintable Primary Writing Paper  PrintableTempl
Printable Primary Writing Paper PrintableTempl
Ā 
Mla Format Template With Cover Page HQ Printabl
Mla Format Template With Cover Page  HQ PrintablMla Format Template With Cover Page  HQ Printabl
Mla Format Template With Cover Page HQ Printabl
Ā 
Reaction Paper Introduction Sample. Reaction Paper In
Reaction Paper Introduction Sample. Reaction Paper InReaction Paper Introduction Sample. Reaction Paper In
Reaction Paper Introduction Sample. Reaction Paper In
Ā 
Persuasive Essay Essay Writing Help Online
Persuasive Essay Essay Writing Help OnlinePersuasive Essay Essay Writing Help Online
Persuasive Essay Essay Writing Help Online
Ā 

Recently uploaded

Proudly South Africa powerpoint Thorisha.pptx
Proudly South Africa powerpoint Thorisha.pptxProudly South Africa powerpoint Thorisha.pptx
Proudly South Africa powerpoint Thorisha.pptxthorishapillay1
Ā 
Framing an Appropriate Research Question 6b9b26d93da94caf993c038d9efcdedb.pdf
Framing an Appropriate Research Question 6b9b26d93da94caf993c038d9efcdedb.pdfFraming an Appropriate Research Question 6b9b26d93da94caf993c038d9efcdedb.pdf
Framing an Appropriate Research Question 6b9b26d93da94caf993c038d9efcdedb.pdfUjwalaBharambe
Ā 
Presiding Officer Training module 2024 lok sabha elections
Presiding Officer Training module 2024 lok sabha electionsPresiding Officer Training module 2024 lok sabha elections
Presiding Officer Training module 2024 lok sabha electionsanshu789521
Ā 
ENGLISH 7_Q4_LESSON 2_ Employing a Variety of Strategies for Effective Interp...
ENGLISH 7_Q4_LESSON 2_ Employing a Variety of Strategies for Effective Interp...ENGLISH 7_Q4_LESSON 2_ Employing a Variety of Strategies for Effective Interp...
ENGLISH 7_Q4_LESSON 2_ Employing a Variety of Strategies for Effective Interp...JhezDiaz1
Ā 
Like-prefer-love -hate+verb+ing & silent letters & citizenship text.pdf
Like-prefer-love -hate+verb+ing & silent letters & citizenship text.pdfLike-prefer-love -hate+verb+ing & silent letters & citizenship text.pdf
Like-prefer-love -hate+verb+ing & silent letters & citizenship text.pdfMr Bounab Samir
Ā 
Final demo Grade 9 for demo Plan dessert.pptx
Final demo Grade 9 for demo Plan dessert.pptxFinal demo Grade 9 for demo Plan dessert.pptx
Final demo Grade 9 for demo Plan dessert.pptxAvyJaneVismanos
Ā 
CELL CYCLE Division Science 8 quarter IV.pptx
CELL CYCLE Division Science 8 quarter IV.pptxCELL CYCLE Division Science 8 quarter IV.pptx
CELL CYCLE Division Science 8 quarter IV.pptxJiesonDelaCerna
Ā 
Procuring digital preservation CAN be quick and painless with our new dynamic...
Procuring digital preservation CAN be quick and painless with our new dynamic...Procuring digital preservation CAN be quick and painless with our new dynamic...
Procuring digital preservation CAN be quick and painless with our new dynamic...Jisc
Ā 
Alper Gobel In Media Res Media Component
Alper Gobel In Media Res Media ComponentAlper Gobel In Media Res Media Component
Alper Gobel In Media Res Media ComponentInMediaRes1
Ā 
Earth Day Presentation wow hello nice great
Earth Day Presentation wow hello nice greatEarth Day Presentation wow hello nice great
Earth Day Presentation wow hello nice greatYousafMalik24
Ā 
Organic Name Reactions for the students and aspirants of Chemistry12th.pptx
Organic Name Reactions  for the students and aspirants of Chemistry12th.pptxOrganic Name Reactions  for the students and aspirants of Chemistry12th.pptx
Organic Name Reactions for the students and aspirants of Chemistry12th.pptxVS Mahajan Coaching Centre
Ā 
ā€œOh GOSH! Reflecting on Hackteria's Collaborative Practices in a Global Do-It...
ā€œOh GOSH! Reflecting on Hackteria's Collaborative Practices in a Global Do-It...ā€œOh GOSH! Reflecting on Hackteria's Collaborative Practices in a Global Do-It...
ā€œOh GOSH! Reflecting on Hackteria's Collaborative Practices in a Global Do-It...Marc Dusseiller Dusjagr
Ā 
Crayon Activity Handout For the Crayon A
Crayon Activity Handout For the Crayon ACrayon Activity Handout For the Crayon A
Crayon Activity Handout For the Crayon AUnboundStockton
Ā 
Introduction to AI in Higher Education_draft.pptx
Introduction to AI in Higher Education_draft.pptxIntroduction to AI in Higher Education_draft.pptx
Introduction to AI in Higher Education_draft.pptxpboyjonauth
Ā 
Meghan Sutherland In Media Res Media Component
Meghan Sutherland In Media Res Media ComponentMeghan Sutherland In Media Res Media Component
Meghan Sutherland In Media Res Media ComponentInMediaRes1
Ā 
Types of Journalistic Writing Grade 8.pptx
Types of Journalistic Writing Grade 8.pptxTypes of Journalistic Writing Grade 8.pptx
Types of Journalistic Writing Grade 8.pptxEyham Joco
Ā 
Computed Fields and api Depends in the Odoo 17
Computed Fields and api Depends in the Odoo 17Computed Fields and api Depends in the Odoo 17
Computed Fields and api Depends in the Odoo 17Celine George
Ā 

Recently uploaded (20)

TataKelola dan KamSiber Kecerdasan Buatan v022.pdf
TataKelola dan KamSiber Kecerdasan Buatan v022.pdfTataKelola dan KamSiber Kecerdasan Buatan v022.pdf
TataKelola dan KamSiber Kecerdasan Buatan v022.pdf
Ā 
Proudly South Africa powerpoint Thorisha.pptx
Proudly South Africa powerpoint Thorisha.pptxProudly South Africa powerpoint Thorisha.pptx
Proudly South Africa powerpoint Thorisha.pptx
Ā 
Model Call Girl in Bikash Puri Delhi reach out to us at šŸ”9953056974šŸ”
Model Call Girl in Bikash Puri  Delhi reach out to us at šŸ”9953056974šŸ”Model Call Girl in Bikash Puri  Delhi reach out to us at šŸ”9953056974šŸ”
Model Call Girl in Bikash Puri Delhi reach out to us at šŸ”9953056974šŸ”
Ā 
Framing an Appropriate Research Question 6b9b26d93da94caf993c038d9efcdedb.pdf
Framing an Appropriate Research Question 6b9b26d93da94caf993c038d9efcdedb.pdfFraming an Appropriate Research Question 6b9b26d93da94caf993c038d9efcdedb.pdf
Framing an Appropriate Research Question 6b9b26d93da94caf993c038d9efcdedb.pdf
Ā 
Presiding Officer Training module 2024 lok sabha elections
Presiding Officer Training module 2024 lok sabha electionsPresiding Officer Training module 2024 lok sabha elections
Presiding Officer Training module 2024 lok sabha elections
Ā 
ENGLISH 7_Q4_LESSON 2_ Employing a Variety of Strategies for Effective Interp...
ENGLISH 7_Q4_LESSON 2_ Employing a Variety of Strategies for Effective Interp...ENGLISH 7_Q4_LESSON 2_ Employing a Variety of Strategies for Effective Interp...
ENGLISH 7_Q4_LESSON 2_ Employing a Variety of Strategies for Effective Interp...
Ā 
Like-prefer-love -hate+verb+ing & silent letters & citizenship text.pdf
Like-prefer-love -hate+verb+ing & silent letters & citizenship text.pdfLike-prefer-love -hate+verb+ing & silent letters & citizenship text.pdf
Like-prefer-love -hate+verb+ing & silent letters & citizenship text.pdf
Ā 
Final demo Grade 9 for demo Plan dessert.pptx
Final demo Grade 9 for demo Plan dessert.pptxFinal demo Grade 9 for demo Plan dessert.pptx
Final demo Grade 9 for demo Plan dessert.pptx
Ā 
CELL CYCLE Division Science 8 quarter IV.pptx
CELL CYCLE Division Science 8 quarter IV.pptxCELL CYCLE Division Science 8 quarter IV.pptx
CELL CYCLE Division Science 8 quarter IV.pptx
Ā 
OS-operating systems- ch04 (Threads) ...
OS-operating systems- ch04 (Threads) ...OS-operating systems- ch04 (Threads) ...
OS-operating systems- ch04 (Threads) ...
Ā 
Procuring digital preservation CAN be quick and painless with our new dynamic...
Procuring digital preservation CAN be quick and painless with our new dynamic...Procuring digital preservation CAN be quick and painless with our new dynamic...
Procuring digital preservation CAN be quick and painless with our new dynamic...
Ā 
Alper Gobel In Media Res Media Component
Alper Gobel In Media Res Media ComponentAlper Gobel In Media Res Media Component
Alper Gobel In Media Res Media Component
Ā 
Earth Day Presentation wow hello nice great
Earth Day Presentation wow hello nice greatEarth Day Presentation wow hello nice great
Earth Day Presentation wow hello nice great
Ā 
Organic Name Reactions for the students and aspirants of Chemistry12th.pptx
Organic Name Reactions  for the students and aspirants of Chemistry12th.pptxOrganic Name Reactions  for the students and aspirants of Chemistry12th.pptx
Organic Name Reactions for the students and aspirants of Chemistry12th.pptx
Ā 
ā€œOh GOSH! Reflecting on Hackteria's Collaborative Practices in a Global Do-It...
ā€œOh GOSH! Reflecting on Hackteria's Collaborative Practices in a Global Do-It...ā€œOh GOSH! Reflecting on Hackteria's Collaborative Practices in a Global Do-It...
ā€œOh GOSH! Reflecting on Hackteria's Collaborative Practices in a Global Do-It...
Ā 
Crayon Activity Handout For the Crayon A
Crayon Activity Handout For the Crayon ACrayon Activity Handout For the Crayon A
Crayon Activity Handout For the Crayon A
Ā 
Introduction to AI in Higher Education_draft.pptx
Introduction to AI in Higher Education_draft.pptxIntroduction to AI in Higher Education_draft.pptx
Introduction to AI in Higher Education_draft.pptx
Ā 
Meghan Sutherland In Media Res Media Component
Meghan Sutherland In Media Res Media ComponentMeghan Sutherland In Media Res Media Component
Meghan Sutherland In Media Res Media Component
Ā 
Types of Journalistic Writing Grade 8.pptx
Types of Journalistic Writing Grade 8.pptxTypes of Journalistic Writing Grade 8.pptx
Types of Journalistic Writing Grade 8.pptx
Ā 
Computed Fields and api Depends in the Odoo 17
Computed Fields and api Depends in the Odoo 17Computed Fields and api Depends in the Odoo 17
Computed Fields and api Depends in the Odoo 17
Ā 

Folksonomy-based lightweight resource annotation metadata schema for personalized hypermedia learning

  • 1. A folksonomy-based lightweight resource annotation metadata schema for personalized hypermedia learning resource delivery Simon Boung-Yew Laua , Chien-Sing Leeb * and Yashwant Prasad Singhc a Faculty of Engineering and Science, Universiti Tunku Abdul Rahman, 53300 Setapak, Kuala Lumpur, Malaysia; b Graduate Institute of Network Learning Technology, National Central University, 300, Jhongda Road, Jhongli City, Taoyuan 32001, Taiwan; c Faculty of Information Technology, Multimedia University, Jalan Multimedia, 63100 Cyberjaya, Malaysia (Received 1 December 2011; ļ¬nal version received 12 August 2012) With the proliferation of social Web applications, users can now collaboratively author, share and access hypermedia learning resources, contributing to richer learning experiences outside formal education. These resources may or may not be educational. However, they can be harnessed for educational purposes by adapting and personalizing them to diļ¬€erent learner needs. We propose to collaboratively annotate learning resources with a lightweight resource annota- tion metadata schema complemented by a folksonomy-derived semantic model. The annotation metadata schema follows a novel policy to associate numerical ratings to learnersā€™ subjective expression of opinions on learning resources. On the other hand, the semantic model serves to support a collaborative resource recommendation algorithm based on the k-nearest neighbour approach. Proof-of- concept is demonstrated via a prototype Web-based recommender. Preliminary results indicate learners are conļ¬dent with the recommended learning resources in terms of accuracy, usefulness, novelty and information adequacy. Formalization of folksonomy in annotating learning resources as well as in opinion mining from the crowd to rate resources may not only support personalization but also engage learners more eļ¬€ectively in technology-enhanced learning. Keywords: social Web; hypermedia learning resources; personalization; folksonomy; resource annotation metadata; collaborative ļ¬ltering; technology- enhanced learning 1. Introduction and motivation The nature of conventional education is top-down where learning resources are mainly prepared and authored by domain experts and instructors. Knowledge is delivered from the instructors to learners and takes the approach of ā€˜ā€˜one size ļ¬ts allā€™ā€™. This research is motivated by a hypothetical learning scenario illustrated as follows: A young adult, Paul in his early 20s does not like to read academic textbooks or attend training classes provided by experts. He would like to self-learn a highly technical topic such as ā€˜ā€˜Personal Financeā€™ā€™ (ļ¬nance) or ā€˜ā€˜Mobile and Satellite Communicationā€™ā€™ *Corresponding author. Email: cslee@cl.ncu.edu.tw Ɠ 2012 Taylor & Francis Interactive Learning Environments, 2015 Vol. 23, No. 1, 79ā€“105, http://dx.doi.org/10.1080/10494820.2012.745429
  • 2. (engineering) new to him without going to school or without the guidance of any instructor. As he has mobile broadband access at his ļ¬ngertips, he intends to source for useful educational resources suitable for a beginner like him on the Internet. He is on social network such as Facebook almost daily. He reckons that he can gain access to relevant learning resources anytime, anywhere and with the help of anyone who shares and tags various Web articles, blog posts, videos, audio podcasts etc. on a social network. Hence, such tagging and sharing activities may be of help to Paul provided he knows where they are (they are easily searchable). Similarly, other users may beneļ¬t from what Paul tags and shares through his daily Web activities. As reported in Shroļ¬€ (2011), the ā€˜ā€˜Y generationā€™ā€™ or ā€˜ā€˜Facebook Generationā€™ā€™ spends an average 5 h a day on social network and Web 2.0 applications. The advent and prevalence of Web 2.0 or ā€˜ā€˜Read-Writeā€™ā€™ Web technologies such as social network, blogs, wikis and social bookmarks among the younger generation has facilitated social interaction and collaboration among users and, hence, opened the door for authoring and sharing of educational hypermedia resources such as Web articles, videos, images and podcasts for personalized self-paced learning on the Web (Ulbrich, Kandpal, & Tochtermann, 2003). Although the increase in the number of resources available in the open Web can be harnessed for educational purposes, they may or may not be ļ¬t for educational purposes for all users. What is evidently necessary for this mode of learning is not only a facility to understand the educational needs of learners, but also a tool to enable learning resources to be annotated for personalization of these resources to diļ¬€erent learner needs. Due to the overwhelmingly huge amount of learning resources generated and available online, we argue that annotation of such resources by a small group of instructional designers or domain experts is not only time-consuming, but also prone to be inaccurate (due to the lack of time interacting with every piece of information and timely updates). We thus propose that user participation in the annotation process would not only reduce the cost of updating and maintaining resource description metadata, but also improve the quality of resource description such as serendipity (Mathes, 2004; Quintarelli, 2005) and better generalization of new knowledge (Swarup & Gasser, 2008) from a larger pool of diverse opinions. To enable resource discovery and adaptation for personalized learning, collaborative ļ¬ltering (CF) is commonly adopted to label, classify, select and ļ¬lter hypermedia resources on the Web. The CF based on attribute description or numerical ratings to enable resource discovery and adaptation for personalized learning is relatively new (Khribi, Jemni, & Nasraoui, 2007, 2009; Itmazi & Megias, 2008; Tan, Guo, & Li, 2008). In the latter type of CF, it builds a model from a community of usersā€™ past behaviour (such as ratings given to items) to predict ratings for items that the user may have an interest in. However, generally, contextual information such as ā€˜ā€˜whoā€™ā€™ and ā€˜ā€˜whyā€™ā€™ a resource is being liked or disliked is not suļ¬ƒciently captured in CF. In addition, each user may perceive and rate a resource from diļ¬€erent perspectives. For instance, a video tutorial which is ranked 5 on a scale of 1 (worst) to 5 (best) by one user may be due to the quality of its content, while the resource may be ranked 5 by another user due to his preference to the way the content is being presented. Both users may also diļ¬€er signiļ¬cantly in the way they learn. To close the gap, we look towards using collaborative tagging to describe learning resources. S.B.-Y. Lau et al. 80
  • 3. Collaborative tagging is the collective action of users associating tags to resources created or experienced by them (Bouillet, Feblowitz, Liu, Ranganathan, & Riabov, 2008; Pfeiļ¬€er & Tonkin, 2009). The classiļ¬cation of these resources using tags produces folksonomy. Folksonomy is spontaneous, non-hierarchical categorization of Web resources using shared keywords by a community of users (Li & Lu, 2008; Vanderwal, 2007). Conceptually, a generic model of collaborative tagging comprises a tripartite three-uniform structure of entities, i.e. the users, tags and resources. Hence, a collaborative tagging system can be represented as a tripartite hypergraph with hyperedges. Each hyperedge will have vertices of three types which represent a user, a resource being tagged and a tag in a folksonomy where the set of vertices on the folksonomy can be partitioned into three disjoint sets of users, U Ā¼ {u1, u2, . . . , uk}, tags, T Ā¼ {t1, t2, . . . , tl} and resources, R Ā¼ {r1, r2, . . . , rm}. A folksonomy network can be visualized as a tripartite hypergraph H Ā¼ V; E Ć° ƞ Ć°1ƞ where V denotes the vertices in the tripartite graph, V Ā¼ U [ T [ R and E denotes the set of hyperedges. Each hyperedge connects exactly three vertices, one from each set of U, T and R where E U 6 T 6 R Ā¼ {(u,t,r) j u 2 U, t 2 T, r 2 R)} is a ternary relation among U, T and R whose elements are instances of annotation or tag assignment by a particular user from U with a tag in T for a resource in R (Mika, 2005; Zhang Chuang, 2010). Hence, folksonomy has great potential in enriching learning resource description metadata as well as automated selection, ļ¬ltering and ranking of learning resources on the open Web for personalization of these resources to learnersā€™ educational needs. To the best of our knowledge, there is a lack of semantic support for social tags for learning in social contexts. Thus, the research problem addressed in this article is how to enable the learning system to make sense of the free, unconstrained and arbitrary tag metadata (Guy Tonkin, 2006). Our work is centred on the research question: How can we formalize the deļ¬nition of user-contributed tag metadata to facilitate personalization? We propose a semantic model for folksonomy-derived resource description metadata and a policy for resource annotation based on the semantic model to formalize semantic information about learning resources as well as users who tag them. This enables resources to be selected based on user context proļ¬le (e.g. learning style and competency level) and educational preferences (e.g. theme, content format and complexity level of resources). The working principle of how the description metadata schema can be integrated into a learning resource recommen- dation approach based on k-nearest neighbour will be presented in Section 8. 2. Contribution The focus of our work is diļ¬€erent from the above-mentioned related work on folksonomy from two perspectives. Firstly, our work focuses on the application of learning resource personalization in informal, self-directed social learning environ- ment while most related work focus on conventional e-learning application in formal education settings. Interactive Learning Environments 81
  • 4. Secondly, as far as CF of learning resources is concerned, there is currently limited formal conceptualization that represents folksonomy in a consistent way (Kim, Breslin, Yang, Kim, 2008) Consistent representation of folksonomy facilitates systematic annotation process. Furthermore, there is no semantic annotation schema representing knowledge about learning resources attained from tags. Hence, we aim to formulate a novel semantic model for tags which serves as a tool and policy framework to associate and correlate usersā€™ subjective tag keywords metadata to high-level concepts depicting features of a learning resource such as theme, content format and complexity level as well as numerical rating of the resource. The model includes a well-deļ¬ned resource rating policy based on user opinion. The semantic model and annotation technique address the issue of formalizing collaborative social tags in informal learning. Besides, a novel domain ontology for ā€˜ā€˜personal ļ¬nanceā€™ā€™ was devised for experimentation because as far as we have surveyed, there is currently no readily deļ¬ned ontology for this domain. Thirdly, another important aspect of learning resource description proļ¬ling is mining the opinions of the user community about learning resources. At present, two popular opinion mining techniques on Web 2.0 systems are the ā€˜ā€˜Likeā€™ā€™ voting on the Facebook social network platform and numerical ratings. These techniques generally suļ¬€er from a weakness: They do not naturally take into consideration ā€˜ā€˜contextā€™ā€™ in which an item is being rated (Nakamoto, Nakajima, Miyazaki, Uemura, 2008). Hence, we also contribute to ļ¬ll this gap by devising a user ā€˜ā€˜opinion-ratingā€™ā€™ policy. This policy is devised to extract the feelings of users about how ā€˜ā€˜goodā€™ā€™ a learning resource is for the purpose of rating and ranking the resources. Lastly, our semantic model provides the framework for (1) item-based ļ¬ltering of learning resources prior to a user-based classiļ¬cation of resources and (2) ranking of learning resources based on numerical ratings inferred from tags within a collaborative resource recommendation algorithm based on the k- nearest neighbour. 3. Related work Learning resource recommendation is at a relatively early stage of research and development (Itmazi Megias, 2008). A major concern among the research fraternity is identifying the requirements of learning recommender systems (Drachsler, Hummel, Koper, 2007; Santos, Baldiris, Boticario1, Restrepo1, Fabregat, 2011; Santos Boticario, 2011) with issues and challenges speciļ¬c to the learning domain (Tang McCalla, 2005). Most current eļ¬€orts on recommender systems are related to the formation of conceptual models, frameworks or architecture (Itmazi Gea; 2006; Itmazi Megias, 2008; Kerkiri, Manitsaris, Mavridou, 2007; Otair Al Hamad, 2005; Tan et al., 2008) focusing on formal learning (Karampiperis Sampson, 2005). The designs are, in essence, top-down and highly structured where learning material and learning path are designed and administered by domain experts to guarantee quality. However, implementation details of these eļ¬€orts are not in abundance for conclusive remarks to be made about the level of success of such applications. Meanwhile, learning recommender systems for informal learning are slowly developing (Drachsler Manouselis, 2009; Soonthornphisaj, Rojsattarat, S.B.-Y. Lau et al. 82
  • 5. Yim-ngam, 2006). However, there is limited documented information about the implementation detail and the level of success of these systems. Overview of the requirements for learning personalization and the types of recommendation approaches are summarized in Figure 1. As far as learning resource description proļ¬ling is concerned, Fok and Ip (2007) designed and constructed personalized education ontology (PEOnto) through an iterative ontology construction process. The PEOnto is made up of ļ¬ve ontologies of vocabularies. These are: the users, the learning content, the curriculum, the pedagogy and the software agents for describing, organizing, retrieving and recommending hypermedia educational resources. Fok and Ip (2007) have since adopted PEOnto in a prototype called personalized instruction planner (PIP). The PIP is implemented to facilitate instructors in annotating learning objects with semantic description. Meanwhile, Jovanovic, GasĢŒevicĢ, and DevedzĢŒicĢ (2009) devised an ontological framework for learning personalization, which includes a learning content structural ontology, a context ontology to specify the pedagogical role of the content, a domain ontology of concepts, a learning path ontology and a user ontology. These ontologies are used to personalize learning content to the usersā€™ domain knowledge, preferences and learning styles in the domain of intelligent information systems. Both research eļ¬€orts have demonstrated the use of ontologies in authoring learning resources in formal education. Other related work on ontology-based recommendation systems driven by similar motivation (Buriano et al., 2006; Cantador, BellogıĢn, Castells, 2008; Costa, Guizzardi, Guizzardi, Filho, 2007; Naudet, Mignon, Lecaque, Hazotte, Groues, 2008) have endeavoured to enhance the semantic description of user preferences and items with concepts and instances deļ¬ned in ontologies. Eļ¬€orts to construct ontologies to represent knowledge and concepts in e-learning include those by Heiyanthuduwage and Karunaratne (2006), SĢŒimuĢn, Andrejko, and BielikovaĢ (2007) and Tsai, Chiu, Lee, and Wang (2006). Some of these eļ¬€orts are summarized in Table 1. These approaches exploit learning personalization based on domain knowledge and user context. Figure 1. Framework for learning resource recommender systems. Interactive Learning Environments 83
  • 6. There are various approaches employed in constructing domain ontologies. Apart from the knowledge engineering approach adopted in this work (Yun, Xu, Wei, Xiong, 2009), there are the iterative convergence model (Pinto Peralta, 2003), composition/integration model (Lin, Tseng, Weng, Lin, Su, 2007), skeletal (Uschold GruĢˆningerm, 1996), seven-step method (Yun et al., 2009) and ļ¬ve-step recipe (Yun et al., 2009), whose phases of development are conceptually similar. In this article, we adopt the widely applied knowledge engineering approach which is generically modelled after the IEEE 1074-2006 Software Development Life Cycle processes and proven eļ¬€ective in the e-learning domain (Yun et al., 2009) to develop the ontology for the personal ļ¬nance domain. More recently, there is growing popularity on leveraging collaborative tags for context clues to supersede CF based on user ratings (Gadepalli, Rundensteiner, Brown, Claypool, 2010; Milicevic, Nanopoulos, Ivanovic, 2010; Nakamoto et al., 2008; Tso-Sutter, Marinho, Schmidt-Thieme, 2008). This initiative is based on the assumption that similar users will use similar or related keywords to express their opinions on similar items. This is also the opinion harvesting approach that our article is presenting. Details of the approach will be presented in Section 6.4. Meanwhile, resource recommendation approaches can be classiļ¬ed based on the way ļ¬ltering and selection of resources are done, i.e. content-based (item similarity) and CF-based (user similarity) or based on the ways in which ratings of the resources are being estimated, i.e. memory-based and model-based approaches. Hybrid approaches combine more than one approach for speciļ¬c purposes. Taxonomy of the resource recommendation approaches is shown in Figure 2. In general, most recommendation approaches in personalized learning systems employ the CF approach compared to the content-based approach due to the presence of the concept of ā€˜ā€˜communityā€™ā€™ in CF, which makes it a natural choice for learning in a social-networked environment (Khribi et al., 2007, 2009). User-based CF such as k-nearest neighbour (k-NN) is the most prevalent approach employed for learning resource recommendation (Itmazi Megias, 2008; Khribi et al., 2009; Tan et al., 2008). For instance, Khribi et al. (2009) classify learners using k-NN to recommend Web links to learners based on cosine similarity. On the other hand, Shih and Lee (2001) employ the k-NN to provide adaptive learning materials for a Table 1. Summary of work which exploits ontologies to represent knowledge and concepts. Semantic technology exploited Personalization Application Heiyanthuduwage and Karunaratne (2006) Ontology that deļ¬nes relationship between metadata for learning objects Maps user preferences with learning content Conventional open sourced learning management system (LMS) SĢŒimuĢn et al. (2007) Ontology of knowledge item space and concept space Adapt learning application to user model Web-based e-learning application Tsai et al. (2006) Ontology of course content Rank the degree of relevance of learning objects to a userā€™s preference and nearest neighboursā€™ preferences Not stated S.B.-Y. Lau et al. 84
  • 7. computer networking course. It is interesting to note that to date, as far as we have surveyed, item-based CF is yet to be employed prevalently in the e-learning domain in any signiļ¬cant way. 4. Outline of the article The rest of this article is organized as follows. Firstly, we conceptualize the formalization of the meanings of tags in keyword metadata in folksonomy. In Section 6, we present the ontology engineering process of a semantic model for tags. In Section 7, we demonstrate how the semantic ontology for tags can be applied in the resource annotation process for a collaborative resource recommendation algorithm based on k-nearest neighbour approach. In Section 8, we present the user evaluation results of a prototype Web-based learning recommender system implementing the said model and algorithm. 5. Formalizing collaborative tags Conventionally, there is no formal conceptualization for representing folksonomy in a consistent way (Kim et al., 2008). There is no systematic process to guide the tagging process and a semantic annotation schema representing the knowledge attained from tags. As far as we know, there is no commonly known popular Web 2.0 tagging system representative of the learning scenario we are targeting, which focuses on educational resources with large enough data size that warrants a conclusive analysis of tags. As far as we know, currently there is no commonly known popular Web 2.0 tagging system on educational resources that has large enough data size which warrants a conclusive analysis of tags. Hence, we select Delicious.com as the sample tag database for our observation. We assume every piece of Web resource can be educational to speciļ¬c segments of audiences in speciļ¬c contextual situations. Our study indicates that folksonomy can be principally classiļ¬ed into categories as follows (Table 2; Marchetti, Tesconi, Ronzano, 2007; Xu, Fu, Mao, Su, 2006): (1) Factual: up to 80% of tags describe the theme of the content being tagged and approximately 17% describe the content format of the resource or the context in which the resource is used; Figure 2. Taxonomy of resource recommendation approaches. Interactive Learning Environments 85
  • 8. (2) Descriptive: about 3% of tags describe usersā€™ subjective emotions or opinions. Though the study stated above clearly indicates the presence of distinct types of semantic information in tag data, it is inherently diļ¬ƒcult to automatically extract the implicit meanings of tag data since there is lack of organization or structure of higher-level semantic relations between tags. A single user ā€˜ā€˜votesā€™ā€™ for a resource with its tag and the votes are counted only as one in standard tagging applications. That is, if tag1 Ā¼ tag2, then there is no diļ¬€erence semantically between the ļ¬rst and subsequent assertions, which means the additional assertions are logically redundant. For the purpose of harnessing tags to describe learning resources, we need to devise a policy to guide the tagging process and organize of the meanings of tags. Simply put, it is to make folksonomy able to ā€˜ā€˜talkā€™ā€™ to a learning system (Li Lu, 2008). We propose to undertake this by constructing a semantic model for tags to address the following semantic information about learning resources: (1) the theme of the subject matter the learning resource is related to: what the resource is about (tags such as ā€˜ā€˜personal ļ¬nanceā€™ā€™ and ā€˜ā€˜mobile communicationā€™ā€™), (2) the content format of the learning resource (tags such as ā€˜ā€˜videoā€™ā€™ and ā€˜ā€˜textā€™ā€™), (3) the quality attributes of the resource perceived by users (tags such as ā€˜ā€˜boringā€™ā€™ and ā€˜ā€˜coolā€™ā€™) and (4) proļ¬le of the tagger who creates and owns the tag. We thus propose formalizing folksonomy with a semantic model for tags as shown in Figure 3 with the following features: (1) a domain ontology that describes the relations of concepts, (2) a content format ontology and (3) an opinion-rating policy to associate numerical rating to subjective opinions of users to grade the learning resources. For proof-of-concept, formulation of the semantic model is presented in detail in the following sections. 6. Semantic model for tags 6.1. Domain ontology In this section, we present the knowledge engineering process to constructing a novel domain ontology on ā€˜ā€˜personal ļ¬nanceā€™ā€™ for illustration purpose. The goal of the Table 2. Breakdown of tags by type of semantic information. Category Type of semantic information Percentage (%) Factual Theme of a resource (content-/context-based tags) 78.06 Content format (the kind of informative content) 17.35 Descriptive Subjective opinion or emotion about the level of ā€˜ā€˜goodnessā€™ā€™ 3.06 Others Example: preposition, conjunction and organizational 1.53 S.B.-Y. Lau et al. 86
  • 9. domain ontology is to provide the basis for organizing the domain themes. The purpose of the domain ontology is to guide the learning resource annotation process to deļ¬ne concepts in the knowledge domain. The domain ontology is extensible with concepts deļ¬ned by users in the form of tags. It will be implemented in the future versions of the prototype system described in Section 9. The knowledge engineering approach to construct the domain ontology is adopted from the IEEE 1074-2006 Software Development Life Cycle standard reported in Yun (2009) and Yun et al. (2009). Phases of the ontology construction process include ontology competence speciļ¬cation, ontology acquisition, ontology implementation and ontology evaluation as shown in Figure 4. Firstly, the goal and scope of the ontology, its intended uses and target end users are deļ¬ned and speciļ¬ed. Competency questions for the ontology are designed through a few iterations of deliberations and reviews to test whether the goals and scope of the ontology have been expressively covered. Whether a concept or axiom is to be included in the domain ontology are veriļ¬ed by the questions. For instance, competency questions for ā€˜ā€˜fundamental concepts in personal ļ¬nanceā€™ā€™ are such as: (1) What are the general principles of personal ļ¬nance? (2) What are the fundamental concepts of personal ļ¬nance for a learner (beginner) to know? (3) What are the key knowledge areas related to personal ļ¬nance? According to the above-mentioned ontology competency questions, relevant concepts, relations, properties and roles are identiļ¬ed and organized into a semantic structure by following these steps (Lau Lee, 2012): Figure 4. Ontology engineering process of the domain ontology. Figure 3. Semantic model for tags. Interactive Learning Environments 87
  • 10. (1) Enumerate important concepts and terms: important concepts and terms for the subject domain are manually extracted from related closed-corpus and open-corpus materials via a careful process of deliberation. Closed-corpus materials include textbooks and reference books (Altfest, 2007; Bajtelsmit, 2006; Downey, 2005; Frasca, 2009; Kapoor, 2009; Michael, 1997) and online educational materials (Blecksmith, 2010; Keown, 2010) authored by domain experts. Open-corpus materials include open Web resources such as wiki or blog posts such as GetRichSlowly.com, MyMoney.com and Wisebread.com, news portals (e.g. CNN Money, Forbes Finance and Yahoo Finance) and DMOZ Open Directory Project (2012) about personal ļ¬nance. The titles, subtitles, outlines, table of content and so on were observed to extract the intrinsic semantic structure of the subject domain. Terms or keywords extracted from these materials form the conceptual model. Apart from the information extraction approach described above, in recent years, it may be also possible to learn the ontologies from the Web using automated tools. However, as far as we know, most of these semantic Web tools are still at the early stage of implementation and there is rarely universally constructed and accepted ontologies on domains such as ā€˜ā€˜personal ļ¬nanceā€™ā€™. An experiment with Swoogle, a semantic Web search engine shows that there is currently no any readily deļ¬ned ontology which is near to the targeted domain in the area of ā€˜ā€˜personal ļ¬nanceā€™ā€™. (2) Deļ¬ne concepts, attributes and relations of concepts and organize hierarchy structure: Concepts in the domain are categorized and organized top-down into hierarchical structures as shown in Figure 5. For this research, the approach proposed by Leo, Ceusters, Mani, Ray, and Smith (2007) is adopted to verify the ontology where the ontology is assessed by domain experts against a set of criteria. The purpose of the evaluation is not only to assess the quality of the ontology, but also to enhance the ontology based on the suggestions of the experts. The quality of the domain ontology was evaluated against a set of 12 competency statements by a total of ļ¬ve subject experts in the Finance Department, Faculty of Management of Multimedia University on a Likert scale from ā€˜ā€˜strongly disagreeā€™ā€™ (1) to ā€˜ā€˜strongly agreeā€™ā€™ (5). The results of the assessment are as shown in Table 3. For the purpose of comparison, the ratings are aggregated into a quality score by summing the product of rating and the number of evaluators. In conclusion, the domain ontology passes all the 12 quality criteria by attaining average scores between 17 and 20 out of 25. Subsequently, the ontology was also validated against the competency questions by the same group of experts. The domain ontology under experiment was unanimously rated as fulļ¬lling all competency questions. 6.2. Hypermedia content format ontology Content format indicates whether the learning resource is a text, image, audio, video, vendor speciļ¬c ļ¬le format, Web service, etc. The hypermedia content format ontology is devised with reference to the deļ¬nition of Internet media type (Grafman Productions, 1997) to guide users to tag learning resources with the right content format. The tags specifying the content format are used to ļ¬lter learning resources. S.B.-Y. Lau et al. 88
  • 11. Figure 5. Sample ontology of fundamental concepts in personal ļ¬nance. Interactive Learning Environments 89
  • 12. As shown in Figure 6, common content formats for hypermedia resources are text, image, audio, video and Web applications. 6.3. Complexity level Apart from the above-mentioned keywords, complexity level of the learning resources can also be tagged with a set of tag keywords, T Ā¼ {ā€˜ā€˜Not Complexā€™ā€™, Table 3. Ontology quality evaluation results. Number of evaluators who rated Score 1 2 3 4 5 No. Criteria Strongly disagree Neutral Strongly agree Score Average Std dev. 1. The ontology suļ¬ƒciently and completely deļ¬nes all the principal high-level concepts/terms in personal ļ¬nance. ā€“ 1 ā€“ 4 ā€“ 18 3.6 2.12 2. The ontology has completely deļ¬ned all the principal relations in personal ļ¬nance and used them in consistent ways. ā€“ 1 ā€“ 4 ā€“ 18 3.6 2.12 3. The use of vocabularies in the ontology is consistent and unambiguous. ā€“ ā€“ ā€“ 5 ā€“ 20 4 ā€“ 4. The ontology is correct and does not require much corrections and/or additions of terms. ā€“ 1 ā€“ 4 ā€“ 18 3.6 2.12 5. The ontology is accurate for its intended use, i.e. for learning and mastering a wide spectrum of concepts in personal ļ¬nance. ā€“ ā€“ 2 3 ā€“ 18 3.6 0.71 6. The ontology is compact and does not have any redundant deļ¬nition of vocabularies or ā€˜ā€˜circular deļ¬nitionsā€™ā€™. ā€“ ā€“ 2 3 ā€“ 18 3.6 0.71 7. The ontology is precise and does not give a new meaning to a term in place of its original vocabulary already established in use in personal ļ¬nance. ā€“ ā€“ 3 2 ā€“ 17 3.4 0.71 8. The ontology is comprehensive and contains suļ¬ƒcient high-level concepts to understand the overall big picture of ā€˜ā€˜personal ļ¬nanceā€™ā€™. ā€“ ā€“ 2 3 ā€“ 18 3.6 0.71 9. The ontology is detailed enough to contain suļ¬ƒcient low-level concepts to have in-depth and comprehensive study of personal ļ¬nance. ā€“ ā€“ 1 4 ā€“ 19 3.8 2.12 10. The ontology contains vocabularies which are formal and standard in the area of personal ļ¬nance. There is small or no informal or ad hoc non-standard terms used. ā€“ ā€“ 1 4 ā€“ 19 3.8 2.12 11. The ontology is easily extensible. It is possible to easily add lower-level concepts into this ontology and extend the knowledge domain with its current structure. ā€“ ā€“ 2 3 ā€“ 18 3.6 0.71 12. The ontology is up to date. It reļ¬‚ects the most current and up-to-date view of the fundamental concepts in personal ļ¬nance. ā€“ ā€“ 2 3 ā€“ 18 3.6 0.71 S.B.-Y. Lau et al. 90
  • 13. ā€˜ā€˜Complexā€™ā€™, ā€˜ā€˜Quite Complexā€™ā€™, ā€˜ā€˜Very Complexā€™ā€™, ā€˜ā€˜Extremely Complexā€™ā€™} which can be translated into numerical values g(t) as shown in Table 4. For instance, if a user thinks that a resource is ā€˜ā€˜Not Complexā€™ā€™ at all, he is indirectly assigning a value of ā€˜ā€˜0ā€™ā€™ to the ā€˜ā€˜Complexityā€™ā€™ ļ¬eld of the resource query. 6.4. Opinion-rating policy The domain ontology and hypermedia content format ontology are used to determine the degree of relevance of learning resources to a userā€™s query. They are not able to capture quality of the resources as perceived by users. To ļ¬ll this gap, a user ā€˜ā€˜opinion-ratingā€™ā€™ policy is devised to extract the opinions of users about how ā€˜ā€˜goodā€™ā€™ a learning resource is for the purpose of rating and ranking the resources. The policy functions to associate commonly expressed subjective opinion keywords of users appearing in tag data to objective numerical rating scores as a product function: Composite rating score; g Ā¼ g d Ć° ƞ g l Ć° ƞ g o Ć° ƞ Ć°2ƞ Table 4. Quantifying complexity level of resources. Tag, t Complexity level, g(t) ā€˜ā€˜Notā€™ā€™ ā€˜ā€˜Complexā€™ā€™ 0 ā€˜ā€˜ā€™ā€™ 1 ā€˜ā€˜Quiteā€™ā€™ 2 ā€˜ā€˜Veryā€™ā€™ 3 ā€˜ā€˜Extremelyā€™ā€™ 4 Figure 6. Content format ontology. Interactive Learning Environments 91
  • 14. where d is the member of a set of keywords expressing intensity of feeling Ā¼ {ā€˜ā€˜Slightlyā€™ā€™, ā€˜ā€˜Moderatelyā€™ā€™, ā€˜ā€˜Veryā€™ā€™, ā€˜ā€˜Extremelyā€™ā€™}, l Ā¼ {ā€˜ā€˜ā€“ā€™ā€™, ā€˜ā€˜notā€™ā€™} and o is the member of a set of opinion keywords such as ā€˜ā€˜coolā€™ā€™, ā€˜ā€˜boringā€™ā€™ and ā€˜ā€˜badā€™ā€™. Each of the terms stated above are translated into corresponding numerical rating score, g Ā¼ g(d) 6 g(l) 6 g(o) as shown in Table 5. For instance, a learning resource tagged as ā€˜ā€˜very boringā€™ā€™ will be rated with rating score, g Ā¼ 4 6 1 6 (71) Ā¼ 7 4 on a scale of 75 to 5 where ā€˜ā€˜veryā€™ā€™ Ā¼ 4, ā€˜ā€˜7ā€™ā€™ Ā¼ 1 and ā€˜ā€˜boringā€™ā€™ Ā¼ 71. The strength of this model is (1) Users are able to rate the learning resources subjectively with phrases expressing their feelings without having to think about the corresponding rating value; (2) Tags are standardized to a standardized system interpretable format. Rating values can converge to values that rank resources without regard to the diverse ways of expressions of feelings by users. 7. Collaborative resource recommendation In a personalized learning process, context metadata (e.g. learning style and competency level) of a learner, and the need and intention of a learner (e.g. the desired theme, content format and complexity level) are elicited indirectly or speciļ¬ed directly by a learner. The query metadata schema is composed of two components, namely the learner context proļ¬le and information need proļ¬le. A query is deļ¬ned as an ordered vector, Q Ā¼ ((Learning Style, Competency Level), (Theme, Content Format, Complexity)). The resource query is the result of a learner context proļ¬ling and a resource annotation process. Schema and policy described of the semantic model for tags described above formalize these processes. As a result, a resource query initializes a learning resource recommendation process as shown in Figure 7. We propose a collaborative resource recommendation algorithm based on k-nearest neighbour with the following stages as shown in Figure 8: Table 5. ā€˜ā€˜Opinion-ratingā€™ā€™ policy for grading the resources. Intensity, d g(d) l g(l) Opinion keywords, o g(o) ā€˜ā€˜Slightlyā€™ā€™ 2 ā€˜ā€˜ā€“ā€™ā€™ 1 Negative ā€˜ā€˜boringā€™ā€™ 71 ā€˜ā€˜Moderatelyā€™ā€™ 3 ā€˜ā€˜toughā€™ā€™ ā€˜ā€˜Veryā€™ā€™ 4 ā€˜ā€˜notā€™ā€™ 71 ā€˜ā€˜badā€™ā€™ ā€˜ā€˜Extremelyā€™ā€™ 5 ā€˜ā€˜sillyā€™ā€™ ā€˜ā€˜lousyā€™ā€™ Ambivalent ā€˜ā€˜no commentā€™ā€™ 0 Positive ā€˜ā€˜coolā€™ā€™ ā€˜ā€˜interestingā€™ā€™ ā€˜ā€˜goodā€™ā€™ 1 ā€˜ā€˜usefulā€™ā€™ ā€˜ā€˜funā€™ā€™ S.B.-Y. Lau et al. 92
  • 15. (1) Stage 1 ā€“ resource matching: the purpose of resource matching is to compute semantic relevance of learning resources based on a learnerā€™s context proļ¬le and pedagogical need speciļ¬ed in a resource query. Resources whose tags fulļ¬l the query of the current active learner are shortlisted. The degree of matching of resources is carried out by computing semantic relevance measures between the learnerā€™s need (preferred theme, content format and complexity level) and features of learning resources as tagged by users based on framework of the semantic model for tags presented in Section 7. The level of matching of a resource follows the function: Smatching Ā¼ a sTheme Ć¾ b sFormat Ć¾ Ć°1 a bƞ sComplexity Ć°3ƞ where Smatching 2 [ 0,1], coeļ¬ƒcient for Theme, b is the coeļ¬ƒcient for Content Format, sTheme is the similarity between themes, sFormat is the similarity between content formats and sComplexity is the similarity between the complexity levels. Resources with Smatching 4 mth, a matching cut-oļ¬€ thresh- old value will be considered to be relevant to a learnerā€™s query shortlisted for next stage of processing. (2) Stage 2 ā€“ resource classiļ¬cation: resources are classiļ¬ed into binary classes: {ā€˜ā€˜Goodā€™ā€™, ā€˜ā€˜Poorā€™ā€™} based on the aggregated similarity measures of context proļ¬les of the current active learner and the learners who tag the resources as well as the average rating inferred from tags. Resources are classiļ¬ed as ā€˜ā€˜Goodā€™ā€™ if the predicted composite rating scores are above a ā€˜ā€˜Goodā€™ā€™ness threshold. Deļ¬nition 1 (ā€˜ā€˜Goodā€™ā€™ or ā€˜ā€˜Poorā€™ā€™ Resources): Resources fulļ¬l the basic criteria of being relevant to the theme, content format and interest of the learner, as well as with high quality content. Technically, ā€˜ā€˜Goodā€™ā€™ness, grtij means the computed rating score is greater than a ā€˜ā€˜Goodā€™ā€™ness threshold value, i.e. grtij 4 gth. The ā€˜ā€˜Goodā€™ā€™ness value is an aggregation of relevance and quality rating of content. Hence, if grtij gth, a resource will be classiļ¬ed as ā€˜ā€˜Poorā€™ā€™. Figure 7. Overview of the personalized learning process. Interactive Learning Environments 93
  • 16. Deļ¬nition 2 (Resource Classiļ¬cation): A resource from a sample set of resources, R Ā¼ {r1, r2, . . . , rk}, is mapped to two classes C Ā¼ {c1, c2} Ā¼ {ā€˜ā€˜Goodā€™ā€™, ā€˜ā€˜Poorā€™ā€™} by a classiļ¬cation function f: ri ! cj 2 C. The classiļ¬ca- tion function is weighted on the similarities between learnersā€™ context proļ¬les: (learning style, competency level). Resources aggregately rated higher than the threshold value, gth by similar learners and other dissimilar learners will be classiļ¬ed as ā€˜ā€˜Goodā€™ā€™, while others are classiļ¬ed as ā€˜ā€˜Poorā€™ā€™. Classiļ¬cation based on a rating function takes into account the similarity of the taggerā€™s context proļ¬les: (learning style, competency level) to the current active learner and the aggregate ratings of the resources weighted by learnersā€™ similarities: Aggregated rating of a resource, grtj Ā¼ 1 N X N iĀ¼1 wi gi Ć°4ƞ Figure 8. Overview of the resource recommendation algorithm. S.B.-Y. Lau et al. 94
  • 17. where N is the total number of rating instances, gi is the rating by ith learner on the resource j, and wi is the weight factor for ri based on learner similarities. Similarity or distance measure between ith learner and the current active learner can be the value for wi to distinguish between the relative signiļ¬cance of ratings imposed by learners. We may also use two distinct values for w where a much larger w, e.g. 100, is used for similar learners (nearest neighbours) compared to smaller w, e.g. 10, for dissimilar learners. (3) Stage 3 ā€“ resource ļ¬ltering: if the rating score computed for a learning resource as grti 4 gth, a resource is considered to be a ā€˜ā€˜Goodā€™ā€™ resource. Else, it is excluded from further consideration as shown in Figure 9. A set of ā€˜ā€˜Goodā€™ā€™ resources, R0 are considered for ranking. (4) Stage 4 ā€“ resource ranking: resources, which are classiļ¬ed as ā€˜ā€˜Goodā€™ā€™ are ranked in the descending order of predicted rating or degree of ā€˜ā€˜goodnessā€™ā€™. (5) Stage 5 ā€“ resource presentation: ranked list of resources are presented to the current active learner for interaction. 8. User study To verify the eļ¬€ectiveness of the semantic model for tags, a prototype Web-based learning resource recommender system has been implemented with PHP-MySQL which is accessible via URLs such as http://3ants.vacau.com. Architecture of the prototype learning resource recommender system is shown in Figure 10. The Figure 9. Resource classiļ¬cation and ļ¬ltering. Interactive Learning Environments 95
  • 18. prototype allows users to share and tag hypermedia resources obtained from the Internet as shown in Figure 11. Users are able to annotate resources with concept(s) or theme(s) for a knowledge domain such as ā€˜ā€˜personal ļ¬nanceā€™ā€™, the content type of the resource and the complexity of the resources. Besides, the key novelty of the prototype system is the subjective opinions expressed by users that can be automatically converted to numerical ratings via the use of commonly found vocabularies of emotion as shown in Figure 12 following an opinion-rating policy described in Section 6.4. The purpose of this pilot user study is aimed at quantifying the various aspects of quality of the recommendation results and the recommender system from the perspectives of the users. As far as we know, currently there is yet any comprehensive framework to evaluate a recommender system for the education domain. Hence, we formulate a novel seven-pillar evaluation framework for learning recommender system as shown in Figure 13. The evaluation framework is formulated via a rigorous process as shown in Figure 14. The system is assessed via questionnaires that encompass evaluation metrics such as user-perceived accuracy, novelty and serendipity, usefulness, domain coverage and diversity of information, conļ¬dence and trust of user, and information adequacy and usability of a learning recommender system. Each of the metrics is explained brieļ¬‚y as follows: (1) User-perceived accuracy is a measure of the level of relevance of the recommender results to what is being expected by users. Though accuracy metrics of recommender systems such as precision, recall and F-measure are more readily measurable via simulated data, it is helpful to conļ¬rm the relevance of the recommended results judged by users. (2) Novelty and serendipity of a recommender system ensure that users have access to non-obvious, unexpected and ā€˜ā€˜surprisingly interestingā€™ā€™ items. Figure 10. Resource sharing and tagging. S.B.-Y. Lau et al. 96
  • 19. (3) Domain coverage or information diversity is a measure of the size of the domain of resources over which the system can make recommendations which may include the content format, themes and topics towards fulļ¬lling the learning needs. (4) It is assumed that higher conļ¬dence and trust in the recommendation would lead to more recommendations being used. (5) Information adequacy is concerned with why an item is recommended as adequately explained and conveyed to the users. (6) Usefulness or eļ¬€ectiveness of the recommendation is concerned with whether the recommendation is suitable to the information needs, learning goal or task performed by the users. (7) Usability is concerned with whether users enjoy the features of the user interface of the recommender system and whether users perceive the tasks as easy to complete. It is more a measure for the recommender system rather than the recommendation results. The pilot user study was carried out as a semi-controlled ļ¬eld study with users in a social Web environment. The user study was piloted for the topic: ā€˜ā€˜Introduction to personal ļ¬nanceā€™ā€™. Invitation to pilot test the prototype system was broadcasted to over 100 direct or indirect friends on the social network. Friends who accepted the Figure 11. Annotating with tags expressing subjective feelings about resource. Interactive Learning Environments 97
  • 20. Figure 12. Architecture of the learning resource recommender system. Figure 13. The seven-pillar evaluation framework. S.B.-Y. Lau et al. 98
  • 21. invitation would log in to the prototype system for pilot runs at a time and place convenient to them over a period of 2 weeks. At the end of the period of experiment, there are a total of 186 resources shared and tagged by a total of 92 friends who accessed the system. However, only about less than half (or 42%) of the respondents completed the learning style and competency-level assessments. The age group of all the respondents is between 20 and 40 with more than half completed or currently pursuing tertiary education. At the end of the process as shown in Figure 15, users went through a 30- question evaluation based on the seven-pillar evaluation framework for the recommended results and recommender system. For reasons unknown, only 11 of the respondents completed the evaluation cycle. The evaluation results are summarized in Table 6. Although the sample size may not be statistically signiļ¬cant at this stage (where only 11 respondents completed the evaluation cycle), the user evaluation results provide useful hints into the factors that aļ¬€ect usersā€™ perception on the quality of recommendation. Preliminary results show that the recommender system and the recommendation results are rated in the positive territory (more than 3.5 from a scale of 1ā€“7) overall, including the relevance of the recommended results (accuracy). The recommendation generally gains conļ¬dence and trust from the users by scoring more than 5 out of 7. Based on Table 6, the recommender is rated between 4.6 and 5 for other quality criteria such as novelty, serendipity, domain coverage and information diversity. As far as we know, as of today, there is no any traditional tagging system speciļ¬cally designed for the educational domain. Eļ¬€ort is underway to build a learning resource recommender system with the traditional tagging approach, and more interesting results will be presented in upcoming publications. 9. Conclusion To sum up, this article has successfully demonstrated the formulation of a lightweight folksonomy-based resource annotation metadata schema based on a semantic model for tags. The framework includes not only a domain that is formally consistent, but is also accurate, extensible and current (based on expertsā€™ assessment). Most importantly, the testing of this framework provides the underlying guiding principles for collaborative annotation of learning resources using semantic tags. The semantic model and metadata schema form essential parts of a personalized learning resource recommendation. The semantic model and Figure 14. Formulation steps of the seven-pillar evaluation framework. Interactive Learning Environments 99
  • 22. Figure 15. Overview of the procedure for the execution of user study. S.B.-Y. Lau et al. 100
  • 23. Table 6. Evaluation results for user study. No. Statement Average rating A Accuracy 1 The items recommended to me matched my needs/preferences. 4.7692 2 The recommended items I received better-ļ¬ts my needs/preferences than what I search using a search engine. 4.8462 3 The recommended results are ranked in the correct order of relatedness to my needs. 4.6923 4 The recommended results are ranked in the correct order of rated quality (in terms of usefulness of the item). 4.6154 Average 4.7308 B Novelty and serendipity 5 The system suggested unexpected new items that are relevant to my needs and preferences. 4.6923 6 The recommender gave me good suggestions. 4.6923 Average 4.6923 C Domain coverage and information diversity 7 The items recommended to me cover diverse content types. 4.2308 8 The items recommended to me cover diverse themes/topics. 4.7692 9 There is enough diversity of items in the results to fulļ¬l my learning needs. 4.6154 Average 4.5385 D Conļ¬dence and trust 10 I trust the recommended items that are ranked correctly. 5.0769 11 I am conļ¬dent in the quality of the items recommended to me. 5.0000 Average 5.0385 E Information adequacy 12 The information provided for the recommended items is suļ¬ƒcient for me. 4.8461 13 I understood the reason why the items were recommended to me. 4.6923 Average 4.7692 F Usefulness and eļ¬€ectiveness 14 The recommender helped me be more eļ¬€ective in learning. 5.0000 15 The recommender helped me be more productive in learning. 5.0000 16 The recommended items inļ¬‚uenced my way of learning. 4.7692 17 I gained new knowledge/skills interacting with the recommended items. 5.0769 18 I am positively stimulated/encouraged by the system to pursue the learning of the subject matter further. 5.0000 19 The system was fun. 4.9231 20 The recommender helped me to decide better. 4.7692 Average 4.9341 G Usability 21 It was easy to look for and access a recommended item(s) I need/like. 4.9231 22 The system was ļ¬‚exible. The recommender provides an adequate way for me to express and revise my needs/preferences. 4.9231 23 I felt comfortable using this system. 4.8461 24 It was easy for me to inform the system if I disliked/liked the recommended item (provide feedback). 4.7692 Average 4.8654 (continued) Interactive Learning Environments 101
  • 24. metadata schema provide the framework for item-based ļ¬ltering of learning resources prior to a user-based classiļ¬cation of resources as well as ranking of learning resources based on numerical ratings inferred from tags within a collaborative resource recommendation algorithm based on the k-nearest neighbour. At present, as a proof-of-concept, the semantic model and metadata schema are implemented in a prototype Web-based learning resource recommender system. Preliminary results of an ongoing user study show that associating collaborative tagging with controlled vocabularies in learning resource annotation is eļ¬€ective in personalizing learning resources. Under the current framework, a learner is able to leverage the opinions of others in order to locate the right, relevant resource of good quality while avoiding less useful ones. Notes on contributors Simon Boung-Yew Lau is a lecturer and researcher at the Faculty of Engineering and Science, Universiti Tunku Abdul Rahman, Kuala Lumpur, Malaysia. He earned his master of science degree from the Malaysia University of Science and Technology. He teaches undergraduate and postgraduate courses, specializing in software engineering. His research areas of interest include context-aware computing, e-learning, social Web and parallel computing. Chien-Sing Leeā€™s research spans across instructional design, computer-supported collaborative learning, game-based learning, mobile learning, data mining and intelligent agents. She has experimented with the use of service-oriented architectures to deliver personalized instruction and the use of knowledge management not only to organize knowledge but also to derive best practices from prior knowledge so that these best practices can be easily adapted to similar scenarios. Another area of interest is interoperability through the semantic Web due to her concern that best practices should be shared, adopted and adapted locally and globally to improve teaching and learning practices in half the time half the cost but with the highest eļ¬€ectiveness and eļ¬ƒciency. Her latest experiments deal with scaļ¬€olding creativity in learning environments. Yashwant Prasad Singh received the B.Sc.Engg. (Electrical) degree (ļ¬rst-class honours) from Bihar College of Engineering, Bhagalpur, India, the M.E. degree in electronics and communications (ļ¬rst-class honours) from the University of Roorkee, Roorkee, India, and the Ph.D. degree in electrical engineering from the Indian Institute of Technology, Kanpur, India, in 1966, 1975 and 1980, respectively. He is currently a professor of computer engineering, Faculty of Information Technology, Multimedia University, Cyberjaya, Table 6. (Continued) No. Statement Average rating H Overall 25 Overall, I am satisļ¬ed with the recommender. 5.0000 26 I am always conļ¬dent I will like the items recommended to me. 4.7692 27 The system was able to convince me about the goodness of the recommendations. 5.0000 Average 4.9231 I Continuance and frequency (future use) 28 I will use this recommender again. 4.7692 29 I will use this recommender frequently. 4.7692 Average 4.7692 J Recommendation to friends 30 I will recommend this recommender to my friends. 5.0000 Overall average 4.8261 S.B.-Y. Lau et al. 102
  • 25. Malaysia. His research interests include complexity theory, cognitive science, computational neuroscience, Artiļ¬cial Neural Network (ANN), fuzzy systems, artiļ¬cial intelligence algorithms and intelligent systems, machine learning and data mining, natural computation (evolutionary computing, neural computation and artiļ¬cial immune systems) and their applications to classiļ¬cation, clustering, anomaly detection, association rule mining, ranking and routing, as well as computer and microprocessor architectures, including transputers. He has published more than 100 papers in international and national journals and conference proceedings. Acknowledgement This work is part of the ļ¬rst authorā€™s PhD research conducted partly in Multimedia University and partly while the second author was a faculty at Multimedia University. References Altfest, L.J. (2007). Personal ļ¬nancial planning. Boston, MA: McGraw-Hill Irwin. Bajtelsmit, V.L. (2006). Personal ļ¬nance: Skills for life. Hoboken, NJ: John Wiley. Blecksmith, R. (2010). Personal ļ¬nance lecture notes continued. Retrieved from http://www. math.niu.edu/*richard/Math101/sp07/monthly_savings_ho.pdf Bouillet, E., Feblowitz, M., Liu, Z., Ranganathan, A., Riabov, A. (2008). A tag-based approach for the design and composition of information processing applications. In Proceedings of the 23rd ACM SIGPLAN conference on object-oriented programming systems languages and applications (pp. 585ā€“602). doi: 10.1145/1449955.1449810 Buriano, L., Marchetti, M., Carmagnola, F., Cena, F., Gena, C., Torre, I. (2006). The role of ontologies in context-aware recommender systems. In The 7th international conference on mobile data management. doi: 10.1109/MDM.2006.149 Cantador, I., BellogıĢn, A., Castells, P. (2008). A multilayer ontology-based hybrid recommendation model. AI Communications, 21(2ā€“3), 203ā€“210. doi: 10.1.1.144.4119. Costa, A., Guizzardi, R., Guizzardi, G., Filho, J. (2007, June). COReS: Context-aware, ontology-based recommender system for service recommendation. Paper presented at the meeting of the 19th International Conference on Advanced Information Systems Engineering, Trondheim, Norway. DMOZ Open Directory Project. (2012). Retrieved from http://www.dmoz.org/ Downey, T. (2005). The Standard Poorā€™s guide to personal ļ¬nance. New York, NY: McGraw-Hill. Drachsler, H., Hummel, H.G.K., Koper, R. (2007)Recommendations for learners are diļ¬€erent: Applying memory-based recommender system techniques to lifelong learning. In The SIRTEL workshop at the EC-TEL 2007 conference (pp. 17ā€“20). Drachsler, H., Manouselis, N. (2009). How recommender systems in technology-enhanced learning depend on context. In The 1st Workshop on Context-aware Recommender Systems for Learning at the Alpine Rendez-Vous 2009. Fok, A.W.P., Ip, H.H.S. (2007). Educational ontologies construction for personalized learning on the web. Studies in Computational Intelligence, 82, 47ā€“82. doi: 10.1007/978-3- 540-71974-8_4 Frasca, R.R. (2009). Personal ļ¬nance: An integrated planning approach. Boston, MA: Prentice Hall. Gadepalli, G., Rundensteiner, E., Brown, D., Claypool, K. (2010). Tag and resource-aware collaborative ļ¬ltering algorithms for resource recommendation. In The 2010 seventh international conference on information technology (pp. 546ā€“551). doi: 10.1109/ITNG. 2010.48 Grafman Productions. (1997). Internet media type. Retrieved from http://graphcomp.com/ info/specs/mime.html Guy, M., Tonkin, E. (2006). Folksonomies: Tidying up tags? Retrieved from http://www. dlib.org/dlib/january06/guy/01guy.html Heiyanthuduwage, S.R., Karunaratne, D.D. (2006). A learner oriented ontology of metadata to improve eļ¬€ectiveness of learning management systems. In The 3rd inter- national conference on eLearning for knowledge-based society, Bangkok, Thailand. doi: 10.1.1.136.8840 Interactive Learning Environments 103
  • 26. Itmazi, J., Gea, M. (2006). The recommendation systems: Types, domains and the ability usage in learning management system. In The international Arab conference on information technology (ACITā€™2006). Itmazi, J., Megias, M. (2008). Using recommender system in course management systems to recommend learning objects. The International Arab Journal of Information Technology, 5(3), 234ā€“240. Jovanovic, J., GasĢŒevicĢ, D., DevedzĢŒicĢ, V. (2009). TANGRAM for personalized learning using the semantic web technologies. Journal of Emerging Technologies in Web Intelligence, 1(1), 6ā€“21. doi: 10.4304/jetwi.1.1.6-21 Kapoor, J.R. (2009). Personal ļ¬nance. Boston, MA: McGraw-Hill Irwin. Karampiperis, P., Sampson, D. (2005). Adaptive learning resources sequencing in educa- tional hypermedia systems. Educational Technology Society. 8, 128ā€“147. doi: 10.1. 1.98.5688 Keown, J. (2010). Financial planning. Retrieved from http://www.ļ¬nance.pamplin.vt.edu/ faculty/ajk/pfm/html/lecture.htm Kerkiri, T., Manitsaris, A., Mavridou, A. (2007). Reputation metadata for recommending personalized e-learning resources. In The second international workshop on semantic media adaptation and personalization (pp. 110ā€“115). doi: 10.1109/SMAP.2007.38 Khribi, M.K., Jemni, M., Nasraoui, O. (2007). Toward a hybrid recommender system for e- learning personalization based on web usage mining techniques and information retrieval. In 2007 world conference on e-learning in corporate, government, healthcare, and higher education (pp. 6136ā€“6145). Khribi, M.K., Jemni, M., Nasraoui, O. (2009). Automatic recommendations for e-learning personalization based on web usage mining techniques and information retrieval. Educational Technology Society, 12(4), 30ā€“42. doi: 10.1109/ICALT.2008.198 Kim, H.-L., Breslin, J.G., Yang, S.-K., Kim, H.-G. (2008). Social semantic cloud of tag: Semantic model for social tagging. In The 2nd KES international symposium on agent and multi-agent systems: Technologies and applications, Incheon, Korea (pp. 83ā€“92). doi: 10.1007/978-3-540-78582-8_9 Lau, B.-Y.-S., Lee, C.S. (2012). Enhancing collaborative ļ¬ltering of learning resources with semantically-enhanced social tags. In The 12th IEEE international conference on advanced learning technologies (ICALT2012) (pp. 281ā€“285). Washington, DC: IEEE. Leo, O., Ceusters, W., Mani, I., Ray S., Smith, B. (2007). The evaluation of ontologies: Toward improved semantic interoperability. Semantic Web: Revolutionizing knowledge discovery in the Life Sciences, 139ā€“158. doi: 10.1.1.82.1024 Li, Q.-F., Lu, S.C.-Y. (2008). Collaborative tagging applications and approaches. IEEE Multimedia, 15(3), 14ā€“21. doi: 10.1109/MMUL.2008.54 Lin, H.-N., Tseng, S.-S., Weng, J.-F., Lin, H.-Y., Su, J.-M. (2007). An iterative, collaborative ontology construction scheme. In The 2nd international conference on innovative computing, information and control (ICICIC 2007) (pp. 150). doi: 10.1109/ICICIC.2007.164 Marchetti, A., Tesconi, M., Ronzano, F. (2007). SemKey: A semantic collaborative tagging system. Paper presented at the meeting of the Workshop, Tagging and Metadata for Social Information Organization, Banļ¬€, Alberta, Canada. Mathes, A. (2004). Folksonomies ā€“ Cooperative classiļ¬cation and communication through shared metadata. Retrieved from http://www.adammathes.com/academic/computer-mediated- communication/folksonomies.html Michael, J. (1997). Personal ļ¬nance on the web: An interactive guide. New York, NY: Wiley. Mika, P. (2005). Ontologies are us: A uniļ¬ed model of social networks and semantics. In International semantic web conference (pp. 522ā€“536). doi: 10.1016/j.websem.2006.11.002 Milicevic, A., Nanopoulos, A., Ivanovic, M. (2010). Social tagging in recommender systems: A survey of the state-of-the-art and possible extensions. Artiļ¬cial Intelligence Review, 33(3), 187ā€“209. doi: 10.1007/s10462-009-9153-2 Nakamoto, R., Nakajima, S., Miyazaki, J., Uemura, S. (2008). Tag-based context- ual collaborative ļ¬ltering. IAENG International Journal of Computer Science, 34(2), 214ā€“219. Naudet, Y., Mignon, S., Lecaque, L., Hazotte, C., Groues, V. (2008). Ontology-based matchmaking approach for context-aware recommendations. In International conference on automated solutions for cross media content and multi-channel distribution (pp. 218ā€“223). IEEE Xplorer Digital Library. doi: 10.1109/AXMEDIS.2008.13 S.B.-Y. Lau et al. 104
  • 27. Otair, M.A., Al Hamad, A.Q. (2005). Expert personalized e-learning recommender system. In The 2005 international conference on e-business and e-learning (EBEL 2005). Pfeiļ¬€er, H., Tonkin, E. (2009). Tagging in context: Information management across community networks. In C. Bouras, V. Poulopoulos, V. Tsogkas (Eds.), Handbook of research on social interaction technologies and collaboration software: Concepts and trends. USA: IGI Global. doi: 10.4018/978-1-60566-368-5.ch015 Pinto, H.S., Peralta, D.N. (2003). Combining ontology engineering subprocesses to build a time ontology. In The 2nd International Conference On Knowledge Capture (pp. 88ā€“95). doi: 10.1145/945645.945660 Quintarelli, E. (2005). Folksonomies: Power to the people. Retrieved from http://www.iskoi. org/doc/folksonomies.htm Santos, O.C., Baldiris, S., Boticario1, J.G., Restrepo1, E.G., Fabregat, R. (2011). Open issues in personalized inclusive learning scenarios. In The international workshop on personalization approaches in learning environment (pp. 54ā€“58). Santos O.C., , Boticario J.G. (2011). Requirements for semantic educational recommender systems in formal e-learning scenarios. Algorithms, 4(2), 131ā€“154. doi: 10.3390/a4030131 Shih, B.-Y., Lee, W.-I. (2001). The application of nearest neighbour algorithm on creating an adaptive on-line learning system. In The 31th ASEE/IEEE frontiers in education conference (pp. 10ā€“13). doi: http://doi.ieeecomputersociety.org/10.1109/FIE.2001.963925 Shroļ¬€, M.P. (2011, July 17). Educating the facebook generation. Daily news analysis. Retrieved from http://www.dnaindia.com/india/column_educating-the-facebook-generation_1566688 SĢŒimuĢn, M., Andrejko, A., BielikovaĢ, M. (2007). Ontology-based models for personalized e- learning environment. Paper presented at the meeting of the 5th International Conference on Emerging e-Learning Technologies and Applications. Soonthornphisaj, N., Rojsattarat, E., Yim-ngam, S. (2006). Smart e-learning using recom- mender system. In Computational Intelligence (pp. 518ā€“523). Berlin: Springer. Swarup, S., Gasser, L. (2008). Collaborative tagging as information foraging (Technical Report UIUCLISā€“2008/1Ć¾ID). Urbana-Champaign: University of Illinois, Graduate School of Library and Information Science. Tan, H.-Y., Guo, J.-F., Li, Y. (2008). E-learning recommendation system. In International conference on computer science and software engineering (pp. 430ā€“433). doi: http://doi.ieee computersociety.org/10.1109/CSSE.2008.305 Tang, T., McCalla, G. (2005). Smart recommendation for an evolving e-learning system: Architecture and experiment. International Journal on E-Learning, 4(1), 105ā€“129. Tsai, K.-H., Chiu, T.-K., Lee, M.-C., Wang, T.I. (2006). A learning objects recommenda- tion model based on the preference and ontological approaches. In The sixth international conference on advanced learning technologies. doi: http://doi.ieeecomputersociety.org/ 10.1109/ICALT.2006.18 Tso-Sutter, K.H.L., Marinho, L.B., Schmidt-Thieme, L. (2008). Tag-aware recommender systems by fusion of collaborative ļ¬ltering algorithms. In The 23rd annual ACM symposium on applied computing (pp. 1995ā€“1999). doi: 10.1145/1363686.1364171 Ulbrich, A., Kandpal, D., Tochtermann, K. (2003). First steps towards personalization concepts in e-learning. In Wissensmanagement (pp. 229ā€“233). Uschold, M., GruĢˆninger, M. (1996). Ontologies: Principles methods and applications. Knowledge Sharing and Review, 2, 93ā€“136. Vanderwal, T. (2007). Folksonomy: Folksonomy coinage and deļ¬nition. Retrieved from http:// vanderwal.net/folksonomy.html Xu,Z.,Fu,Y.,Mao,J.,Su,D.(2006).Towardsthesemanticweb:Collaborativetagsuggestions.Paper presented at the meeting of the Collaborative Web Tagging Workshop, Edinburgh, Scotland. Yun, H.-Y. (2009). Research on building ocean domain ontology. In The Workshop on Computer Science and Engineering, Vol. 1 (pp. 146ā€“150). Washington, DC: IEEE Computer Society. doi: 10.1109/WCSE.2009.641 Yun, H.-Y., Xu, J.-L., Wei, M.-J., Xiong, J. (2009). Development of domain ontology for e- learning course. In 2009 IEEE international symposium on IT in medicine and education (ITME2009) (pp. 501ā€“506). doi: 10.1109/ITIME.2009.5236370 Zhang, Z.K., Chuang, L. (2010). A hypergraph model of social tagging networks. Journal of Statistical Mechanics: Theory and Experiment, 10, 2ā€“14. doi: 10.1088/1742ā€“5468/2010/10/ P10005 Interactive Learning Environments 105
  • 28. Copyright of Interactive Learning Environments is the property of Routledge and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use.