Resource-Based Learning KOM – Multimedia Communications Lab 2Thursday, May 19, 2011 Resource-based learning (RBL) means learning by using resources found on the Web such as on blogs, YouTube, Wikipedia, in forums or searching via Google. This form of learning started about 10 years ago. Even school pupils are now asked to prepare a homework using information found on the Internet. As we can see, search is an important part of RBL.
Types of Search Behavior A. Broder, 2002: A Taxonomy of Web Search Informational Navigational Transactional 59.3% 28% 12.7% KOM – Multimedia Communications Lab 3Thursday, May 19, 2011 Search on the Web has been classified by Information Management scientists. For example, Broder in 2002 classifies search as Transactional search, where a user goes directly to a website like Amazon and buys a book. Navigational like when a user goes directly to a known website like Spiegel Online to read the news. Informational search however is more a browsing or a search for information like googling to find out the meaning of folksonomy. The most often performed search is informational search and this fits to the RBL form of searching.
Social Search Model B.M.Evans & E.H.Chi, 2009: An elaborated model on social search Gather Social Interactions Before Search Requirements Formulate Social Interactions Representation During Search Foraging Social Interactions Social Interactions Sensemaking After Search Organize Distribute Do Nothing Social Interactions KOM – Multimedia Communications Lab 4Thursday, May 19, 2011 The social search model from Evans and Chi describes the search process not as a solitary act but rather as a social process. In the Before Search stage, the reasons for the search and what to be searched for are determined. For example, a school pupil gets the assignment to present a talk on a topic in the news. The pupil probably goes on SchülerVZ and asks others in her class what they will be presenting and asks where they are searching for information. During the search, the user browses several web pages going from one page on to the other and trying to find information which is relevant to the topic...this is called foraging and sensemaking. In the after search phase, the user saves the found pages and can send these by email to others or prints these out or hopes to use these later.
Gather Before Search Create Activity Tree Requirements Text Messages Chat Form Groups Formulate Representation During Search Add Resources to Activity Foraging Recommendations Tagging Sensemaking After Search Browse Knowledge Network Text Messages Chat Activity Experiences Organize Distribute Do Nothing Recommendations Tagging KOM – Multimedia Communications Lab 5Thursday, May 19, 2011 CROKODIL is a BMBF project with the aim to provide support for RBL in a learning community. I have mapped the activities in CROKODIL to the social search model to classify the features offered. In the Before Search stage, the learners can create an activity, like the assignment to prepare a presentation, groups are formed having this common assignment or having similar tasks. The interaction here is supported on the platform with a chat and text messages. During the search, resources found on the Web are attached to the activities created before the search, recommendations are also made by the system informing about similar resources already stored by others on the platform. The learner can also give tags to these resources to use these to classify the resource in order to find the resource later on. After the search, the learner can browse the knowledge network and see what others have stored, find what other resources have the same tags...add new tags to the resources, recommendations are made showing similar resources from other activities or users. Finally, the learner writes about her experiences when working on the activity as a form of reflection when the activity is completed. In this stage, the learner can again contact other learners via the chat or text message functionality on the platform.
KOM – Multimedia Communications Lab 6Thursday, May 19, 2011 CROKODIL has a platform, where learners can set up a profile, set up friends or groups, create activities, view their resources and read their text messages. The chat function is in the upper right hand corner and the tags used are shown down in the right hand lower corner. The platform is used for the before and after stages of searching.
KOM – Multimedia Communications Lab 7Thursday, May 19, 2011 During search, CROKODIL offers a Firefox addon which is installed in the browser and shows the activity tree. A section of text on the website is highlighted and pulled across to the addon, where it is attached to the activity which is currently activated. When saving, tags can be attached to the resources.
Semantic Tagging New s Nu clear April Energy 2011 Japan KOM – Multimedia Communications Lab 8Thursday, May 19, 2011 In CROKODIL, several concepts are implemented such as the pedagogical concept with the activities and the reflection of the experiences made. We also have implemented the concept of semantic tagging. For example I am a learner in CROKODIL, I find a news article about nuclear energy and tag this with News, I read this in April 2011, the disaster happened in Japan and it’s about a nuclear energy disaster.
Personal Resource Network my resource Resource my tag has tag Nuclear pril TagA Energy 1 New 201 Japan s KOM – Multimedia Communications Lab 9Thursday, May 19, 2011 This forms a structure: me, a learner, having a resource, having my tags which tag my resource. This forms a personal resource network or a personal knowledge network.
Personal Resource Network (Personomy) my resource Resource my tag has tag Tag KOM – Multimedia Communications Lab 10Thursday, May 19, 2011 Viewing this abstractly, this forms a graphical structure...objects form the nodes and the relationships the edges between the nodes. This is known as a personomy.
Semantic Tagging?! my resource Resource my tag has tag Nuclear pril TagA Energy 1 New 201 Japan s KOM – Multimedia Communications Lab 11Thursday, May 19, 2011 So why do we call this semantic tagging? I have shown here the tags having different colors, this shows a sort of classification...Nuclear energy describes the topic of the news article, it is a news article as the green tag shows, the location of the nuclear energy disaster was in Japan and this event happened around April 2011
CROKODIL Tag Types Person Tag has tag has tag Event Tag has tag Tag Resource has tag Topic Tag has tag has tag Location Tag Type Tag KOM – Multimedia Communications Lab 12Thursday, May 19, 2011 In CROKODIL, we have these 6 tag types, the generic tag having no type, a location tag, a type tag, a topic tag, an event tag and a person tag for example Angela Merkel.
Community Resource Network owns tag has friend owns resource Resource Wikipedia Type Tag owns resource owns tag has tag has tag kodil has tag Resource Cro Topic Tag Pr oject KOM – Multimedia Communications Lab 13Thursday, May 19, 2011 So back to the graphical structure which gets built...when several learners are on the platform and they store resources and tag them, resources could be connected by the same tag. This is called a community resource network or Community knowledge network.
Community Resource Network (Folksonomy) owns tag has friend owns resource Resource Type Tag owns resource owns tag has tag has tag has tag Resource Topic Tag KOM – Multimedia Communications Lab 14Thursday, May 19, 2011 This builds such a graphical structure which combines individual personomies to build a folksonomy.
CROKODIL Groups and Activities belongs to has activity Participate in Group Activity GKEL Activities has sub-activity Prepare GKEL Sub- PreActivity n sentatio belongs to has resource has activity Resource owns resource KOM – Multimedia Communications Lab 15Thursday, May 19, 2011 In CROKODIL we additionally have groups and activities. For example the activity GKEL activities belongs to the group GKEL and I just prepared a GKEL presentation.
CROKODIL Groups and Activities belongs to has activity Group Activity has sub-activity Sub- Activity belongs to has resource has activity Resource owns resource KOM – Multimedia Communications Lab 16Thursday, May 19, 2011 Abstractly this can be viewed as a folksonomy with additional social relationships and activities to give pedagogical structure to the knowledge network. With these structures, information about other learners or connected activities can be inferred by traversing the relationships between nodes. One challenge here is identifying what is important especially when the network grows to a larger network over time. Another challenge is finding connections between sub-graphs where no structural connection exists.
Recommender Systems KOM – Multimedia Communications Lab 17Thursday, May 19, 2011 We all know recommender systems such as in Amazon. Recommendations help to identify what is important - as well as what could be new and interesting
Recommender Systems Collaborative-Filtering Structure-Based Activity has sub-activity belongs to has activity Sub- Group Activity my group belongs to my activity owns tag my friend my resource Resource Type Tag owns resource my tag has tag Hybrid has tag has tag Resource Topic Tag Knowledge-Based Content-Based Furniture Chair KOM – Multimedia Communications LabThursday, May 19, 2011 There are different kinds of recommender systems. Most systems combine different approaches in hybrid systems. Collaborative-filtering uses the ratings of users on the platform to determine which resources or tags are important or most used or known. Structure based recommender systems we just saw when the relationships between nodes are used to traverse the graph to find related nodes. Content based methods compare the content of the web pages to determine the similarity between them. Knowledge based methods use other sources of information to determine relationships between objects. Such as Furniture an Chair may not be connected but via Wikipedia, the category Furniture-Chair is determined and the connection made.
What is recommended in CROKODIL? Nuclear pril A Tags Energy 1 New 201 Resources Japan s Friends Participate in GKEL Activities Groups Activities Prepare GKEL Presentation KOM – Multimedia Communications Lab 19Thursday, May 19, 2011 In CROKODIL, tags and resources, friends,groups and activities can be recommended.
CROKODIL Activity Ratings Participate in Activity GKEL Activities has sub-activity Prepare GKEL Sub- Resource has resource Presentation Activity has resource my activity Resource my resource KOM – Multimedia Communications Lab 20Thursday, May 19, 2011 I will be focusing on introducing collaborative filtering to CROKODIL by introducing rating of resources in the context of a CROKODIL activity. For example, I am preparing a new presentation and I see this picture has been rated with 5 stars and this presentation with 4 stars because maybe the picture was used by others but the presentation not.
Ranking Recommendation Results Activity has sub-activity belongs to has activity Sub- Group Activity my group belongs to my activity owns tag my friend my resource Resource Type Tag owns resource my tag 1. has tag has tag has tag Resource Topic Tag Recommendations Furniture 2. Chair KOM – Multimedia Communications Lab 21Thursday, May 19, 2011 In general all recommendation methods weight the resources and offer suggestions of how these resources could be ranked. For example according to the ratings, the picture is ranked higher than the presentation.
CROKODIL Features Nuclear pril Groups A Energy 1 New 201 Tag Types Japan s Social Relations Participate in GKEL Activities Activity Tree Prepare GKEL Presentation Activity Ratings KOM – Multimedia Communications Lab 22Thursday, May 19, 2011 In CROKODIL, it will be interesting when ranking to consider the tag types, for example giving the tag type topic more weight than the tag type type or event or location, then one could consider the social relationships to determine what to prioritize, for example common friends or groups. Also interesting will be the information the hierarchical structure the activities present as well as the ratings in the context of these resources.
Explaining Recommendations ahh... now I get it! KOM – Multimedia Communications Lab 23Thursday, May 19, 2011 My next idea is explaining recommendations. This helps the user to understand the system better and helps to build trust.
Explaining Recommendations Most of your friends love this picture! This presentation was also used in your “Introduction to CROKODIL” activity. KOM – Multimedia Communications Lab 24Thursday, May 19, 2011 The learner reflects on these explanations and learns about relationships that were not obvious before. For example, that most of my friends like this picture...now the picture might be more interesting to me or reminding me that I have already made a presentation introducing CROKODIL...maybe I could use this again.
User Feedback to Recommendations Most of your friends love this picture! This presentation was also used in your “Introduction to CROKODIL” activity. KOM – Multimedia Communications Lab 25Thursday, May 19, 2011 Explanations also motivate the learner to give feedback, for example I do not find the picture relevant to my new presentation even though my friends find this picture good. I rather find the old presentation as interesting.
Feedback Loop 1. Rank 2. Explain Feedback KOM – Multimedia Communications Lab 26Thursday, May 19, 2011 The system ranks the resources and suggests the first two selections, the user is given explanations for this and the feedback is received.
Feedback Loop 1. Rank 2. Explain Feedback KOM – Multimedia Communications Lab 27Thursday, May 19, 2011 The feedback now influences the recommendation algorithms and I realize you prefer the presentation, this tells the recommendation system to maybe recommend similar presentations belonging to you rather than popular resources from your friends.
Research Questions How to exploit social relationships between learners to improve personalized recommendations? How to exploit folksonomy structures for recommendation algorithms? How to generate explanations for collaborative and structural recommendations? KOM – Multimedia Communications Lab 28Thursday, May 19, 2011 I plan to investigate these three questions. How to use the social relationships in CROKODIL, the friendships and groups. As well as the graphical structure especially considering the special features in CROKODIL such as the activities, the ratings, the tag types. Then how best to generate explanations using this structure and the ratings or tags.
Summary Resource Based Social Search Learning Model Personal CROKODIL Resource Network Recommendation Systems Collaboraive Community Filtering Resource Semantic Network Ranking Tagging Recommendation Results Explanations Activity Ratings Tag Types Feedback Loop KOM – Multimedia Communications Lab 29Thursday, May 19, 2011 A mind map as summary
Questions & Contact KOM – Multimedia Communications Lab 30Thursday, May 19, 2011 Any questions?