4. The aim of the research is to analyse and
propose a model of intelligent
personalised multi-agent learning system
applying educational data mining methods
and techniques.
This system is modelled based on
original methodology to personalise
learning applying intelligent technologies.
6. Fig. 1. Search history in Clarivate Analytics Web of Science
database
The intention of the systematic review was to answer
the question: “What are the latest contributions to
application of intelligent agents in education?”
7. Based on systematic review, we can summarise that application of intelligent software
agents and multi-agent systems in education has been actively evolving for past two
decades. In two past years (2014-2016), software agents were studied and deployed
to solve a wide array of educational challenges.
This demonstrates that intelligent agents are a promising and powerful way to
personalise learning. Intelligent agents can adapt learning material to different learning
styles of students and leverage innate predispositions for knowledge acquisition on
intellectual, sensory and emotional level.
On the other hand, no research studies were found that have analysed personalised
intelligent multi-agent systems based on creating learners’ profiles based on their
learning styles, and creating ‘optimal’ (in terms of suitability to particular learner’s
profile) learning units.
Therefore, this approach has to be analysed, and appropriate multi-agent system
should be modelled to be designed and piloted in real pedagogical situations.
9. In personalised learning, first of all, integrated learner profiles (models) should
be implemented.
Dedicated psychological questionnaires like Soloman-Felder Index of Learning
Styles Questionnaire should be applied here.
After that, open learning style model should be created, implicit (dynamic)
learning style modelling method should be used, and, finally, the rest features in
the student profile (knowledge, interests, cognitive traits, goals, learning
behavioural type etc.) should be added to the profile.
Ontologies-based personalised recommender system should be created to
suggest learning components suitable to particular learners according to their
profiles. Recommender system should form the preference lists of the learning
components according to the expert evaluation results.
10. Probabilistic suitability indexes should be identified for all learning components in terms of
their suitability level to particular learners.
The method is based on students’ probabilistic learning styles and expert evaluation of
suitability of different learning components to students’ learning styles. Probabilistic
suitability indexes could be calculated for all learning components and all students if one
should multiply learning components’ suitability ratings obtained while the experts evaluate
suitability of the learning components to particular learning styles by probabilities of
particular students’ learning styles.
All learning components in the recommender system should be linked to particular
students according to their probabilistic suitability indexes. The higher the suitability index
the better the learning component fits particular student’s needs.
These suitability indexes should be included in the recommender system, and all learning
components should be linked to particular students according to these suitability indexes.
11. LA / EDM methods and techniques should be used to analyse behaviour of students in e-
learning systems. Acquired this way data may differ from self-reported psychological evaluation
from questionnaires.
Also, potential for knowledge about a student would constantly increase as they learn, do
exercises, take tests and otherwise interact with the educational system. This constant stream of
information should be used to continuously improve students’ models and as consequence the
service provided. Presently, researchers are addressing questions of cognition, metacognition,
motivation, affect, language, social discourse, etc. using data from virtual learning environments,
intelligent tutoring systems, massive open online courses, educational games and simulations,
and discussion forums.
A hybrid learning style identification can cluster learning styles into three or four combinations
based on learning performance, which suggests that the LA / EDM technique can identify
multiple learning styles and problem-solving approaches. Such incorporation of LA / EDM agent
would create new ways of understanding trends and behaviours in students that can be used to
improve learning design, strengthen student retention, provide early warning signals concerning
individual students and help to further personalise the learner’s experience.
12. The basic LA / EDM techniques applicable in this case should be (but not limited for):
(1) Classification to classify each item in a set of data into one of predefined set of
learners group;
(2) Clustering to determine groups of students that need special course profiling;
(3) Association rules to discover interesting relations between course elements which
were used by particular students;
(4) Prediction to predict dependencies of using learning environment’s activities / tools
and final student’s learning outcomes;
(5) Decision Tree of students’ actions.
To determine and to set appropriate algorithm to a new data set is a difficult task because
there is no single classificatory which equally well suited for all data sets. In practice it is
very important to choose the proper classification / clustering or other algorithm to a
particular data set.
14. Fig. 2. Conceptual model of personalised multi-agent intelligent learning system
15. Proposed model is based on consequent application of 5 intelligent software agents:
1. Dedicated psychological questionnaire like Soloman-Felder Index of Learning Styles Questionnaire
should be applied to obtain probabilistic learning styles of the students. Thus, learning styles identification
software agent (1) should be created to obtain these values of students’ learning styles. All probabilistic
learning styles combinations should be stored for each student in his / her profile (model).
2. Open dynamic learner profile creation software agent (2) should be used to create learners’ profiles
(models) using the results obtained by the agent (1) and adding the other features of learners (knowledge,
interests, goals, cognitive traits, learning behavioural type, etc.).
3. Pedagogical suitability software agent should be created to implement recommender system. This agent
should use high-quality vocabularies of learning components and results of the expert evaluation of
suitability of particular learning components to students’ learning styles. This pedagogical suitability
software agent should link optimal learning components to particular students according to probabilistic
suitability indexes. All learning components in the recommender system should be linked to particular
students according to their probabilistic suitability indexes. The higher probabilistic suitability index the
better the learning component fits particular student’s needs.
16. 4. Optimal learning unit / scenario (i.e. learning unit / scenario of the highest
quality) for particular student means a methodological sequence of learning
components having the highest suitability indexes. Thus, the fourth agent for
composition of optimal learning units / scenarios should be created for particular
learners according to their learning characteristics.
These optimal learning units / scenarios should be created by an intelligent
software agent combining learning components optimal for particular learners.
The number of different combinations of learning components that are optimal
for a particular student should be further analysed by the teacher in order to
create and use the learning unit / scenario as pedagogically sound sequence of the
learning components.
For this purpose, an additional ontology linking learning components (learning
objects, activities and environment) according to their mutual suitability should
be created and implemented in the agent.
17. 5. The data on real students’ behaviours in learning environment obtained by
using LA / EDM methods and tools should be used to correct their profiles
according to the data obtained. Thus, LA / EDM software agent should be created
to correct students’ profiles according to their behaviour in the learning
environment implementing recommended learning units.
The wide range of data about behaviour of students should be used to generate
good quality, real-time predictions about suitable material and activities and
success in acquiring knowledge and skills.
Students and teachers should be able to plan their work on the basis of reliable
tools that can produce detailed and personalised recommendations about what
should be done to achieve the best learning outcomes.
Students practically use some learning objects and activities in real learning
practice in learning environment before identifying appropriate probabilistic
suitability indexes and recommending suitable learning units. Here we could
hypothesise that students preferred to practically use particular learning objects
and activities that fit their learning needs mostly.
Using appropriate LA / EDM methods and techniques, it would be helpful to
analyse what particular learning objects and activities were practically used by
these students in the learning environment, and to what extent.
19. In order to create personalised multi-agent intelligent learning system, first of all,
students’ learning styles should be identified using e.g. Felder-Silverman Learning
Styles Model, then creating the full open dynamic learner’s profiles, identifying
probabilistic suitability indexes and recommending personalised learning units for
particular students as well as verifying these learning units using learning
analytics / educational data mining methods and techniques.
Personalised intelligent learning system should be based on multi-agent approach.
It should consist of at least five intelligent software agents:
(1)learning styles identification software agent,
(2)learner profile creation software agent,
(3)pedagogical suitability software agent,
(4)optimal learning units / scenarios creation software agent, and
(5)learning analytics / educational data mining software agent.
20. Basic learning analytics / educational data mining techniques
applicable in this case should be (but not limited for):
(1)Classification to classify each item in a set of data into one of
predefined set of learners group;
(2)Clustering to determine groups of students that need special
course profiling;
(3)Association rules to discover interesting relations between
course elements which were used by particular students;
(4)Prediction to predict dependencies of using learning
environment’s activities / tools and final student’s learning
outcomes; and
(5)Decision tree of students’ actions.
21. Thank you for your attention !!!
Questions ?
Eugenijus Kurilovas
Vilnius University
Vilnius Gediminas Technical University, Professor