Toward a fully automatic learner model based on web usage mining - - PowerPoint PPT Presentation

Toward a fully automatic learner model based on web usage mining with respect to educational preferences and learning styles

Mohamed Koutheaïr KHRIBI University of Tunis mk.khribi@uvt.rnu.tn

LaTICE, University of Tunis ICALT 2013, Beijing, China.

Mohamed JEMNI University of Tunis mohamed.jemni@fst.rnu.tn Olfa NASRAOUI University of Louisville

• lfa.nasraoui@louisville.edu

Sabine GRAF Athabasca University sabineg@athabascau.ca Kinshuk Athabasca University kinshuk@athabascau.ca

2

• Advantages of automatic learner modeling :

 No additional work for learners ;  Uses information from a time span; higher tolerance  Allows dynamic updating of information

ICALT 2013, Beijing, China 15/07/2013

Learner Modelling Approaches Collaborative Automatic

3

Recommendation

(content, collaborative, hybrid)

LO Features Learner Features

LO LOi LO LOn LO LO2 LO LO1

(Learners) (Learning Objects)

1 2 3

Learner- Object LO Interest Measures (Past& Recent) Implicit Usage Data

Learner Modeling Content Modeling

ICALT 2013, Beijing, China 15/07/2013

General Aim

Building an automatic recommendation system for Learning Management Systems.

4

Research Question

How to automatically model learners and groups of learners based on implicit data from their interactions and online activities in the learning management system, taking into account the learners’ * educational preferences * learning styles

ICALT 2013, Beijing, China 15/07/2013

5

General Features of the Proposed Approach

 Automatic, dynamic and based solely on learner usage sessions ;  Input data is composed from collected implicit data tracked and saved in LMS database and/or web server log files ;  Based on web usage mining techniques ;  Educational preferences and learning styles are considered and identified automatically.

ICALT 2013, Beijing, China 15/07/2013

6

 Proposed Learner model components :

LMi = {PRi, LKi, LEPi}

PRi : ith Learner Profile

General student information such as Identification data, Demographic information, etc.

LKi : ith Learner’s Knowledge Component

Sequences of weighted visited learning objects i.e. vectors of visited learning objects or curriculum elements in which the student was interested (learner’s knowledge)

LEPi : ith Learner’s Educational Component

Learner’s educational preferences and Learning style.

Proposed Learner Model

ICALT 2013, Beijing, China 15/07/2013

7

Learner’s Knowledge Component

ICALT 2013, Beijing, China 15/07/2013

8

Learner’s Knowledge Component

ICALT 2013, Beijing, China 15/07/2013

9

LO1 LO2 LO3 LO4 … LOn

LKi =

Learner’s Knowledge Component

The learner’s knowledge component LKi can be represented as a matrix M(p, n), where p is the total number of learner’s sessions and n the cardinality of unique visited learning objects.

ICALT 2013, Beijing, China 15/07/2013

10

Learner’s Educational Component

Composed of the learner’s preferences among visited learning objects and his/her learning style. Detection of the learner’s preferences: What kind of learning object does a learner prefer? Learning objects available in LMS are characterized by many attributes (e.g. format), each of which may have several values (e.g. for format: text, image, video, etc.) that could be preferred or not by the learner. The preferences of a learner i upon these values can be represented, as a vector of interest measures :

Attributes Related values Type_LO(Learning object type) {Resource, Activity} Shape_LO(Learning object shape, if Type_LO = Activity) {Exercise, Simulation, Questionnaire, Assessment, Forum, Chat, Wiki, Assignment} Format_LO(Learning object format, if Format_LO = Resource) {Text, HTML, Image, Sound, Video}

ICALT 2013, Beijing, China 15/07/2013

11

Learner’s Educational Component

ICALT 2013, Beijing, China 15/07/2013

12

Attributes and corresponding values Interest measures Type_LO

Resource LOIM_Type_LOi

Resource

Activity LOIM_Type_LOi

Activity

Shape_LO

Exercice LOIM_Shape_LOi

Exercice

Questionnaire LOIM_Shape_LOi

Questionnaire

test LOIM_Shape_LOi

Test

Forum LOIM_Shape_LOi

Forum

Chat LOIM_Shape_LOi

Chat

Wiki LOIM_Shape_LOi

Wiki

Navigation LOIM_Shape_LOi

Navigation

example LOIM_Shape_LOi

Exeamle

Format_LO

Text LOIM_Format_LOi

Text

Html LOIM_Format_LOi

Html

Image LOIM_Format_LOi

Image

Audio LOIM_Format_LOi

Audio

Vidéo LOIM_Format_LOi

Video

Learner’s Educational Component

ICALT 2013, Beijing, China 15/07/2013

13 Actifve/Reflective Sensing/Intuitive Visual/Verbal Sequential/Global

F S L S M

Content_visit(-) Content_stay(-) Outline_stay(-) Selfass_stay(-) …. ….

Active/Reflective

Example_visit(+) Example_stay(+) Selfass_visit(+) Selfass_stay(+) …. ….

Sensing/Intuitive

Ques_text(-) Forum_visit(-) Forum_stay(-) Forum_post(-) …. ….

Visual/Verbal

Outline_stay(-) Ques_detail(+) Ques_overview(-) Ques_interpret(-) …. ….

Sequential/Global

Commonly incorporated features in LMS

• Content
• Outline
• Example
• Self-Assessment
• Exercice
• Forum
• Navigation

Learner’s Educational Component

 Detection of the learning style (Graf et al., 2008)

ICALT 2013, Beijing, China 15/07/2013

14

LPi , , {(LEP1, .. , LEPp), LSi}

Once learner models are built, we apply a hierarchical multi- level model based collaborative filtering approach on these models, in order to assign learners with common preferences and interests to the same groups, so that feedback from one learner can serve as a guideline for information delivery to the other learners within the same group.

Group Modeling

ICALT 2013, Beijing, China 15/07/2013

Learning Styles Educational Preferences

…

Level1 : Classification  Level2 : Clustering

…

Group vectors

W1 W2 W3 . . . Wi .. Wm

…

LS1 LS2 LS8

… …

 Level3 : Clustering

… … … …

15 ICALT 2013, Beijing, China 15/07/2013

16

 The proposed approach has been implemented and an experiment was performed as part of a recommendation approach  Recommendations are computed with respect to the learner’s clickstreams, his/her learning style and educational preferences, as well as exploiting similarities and dissimilarities among the learner models and educational content.  Moodle LMS is used to implement the proposed approach. We used an online hybrid course (C2i) with 704 learners. Tracked data was successfully extracted from various Moodle tables (primarily from mdl_log table).

Implementation and Proof of Concept Evaluation

ICALT 2013, Beijing, China 15/07/2013

17

Output

component 1

Implicit Query Extractor

Cp1.2 Terms’ Vector Builder Fenêtre glissante (K termes pertinents) Cp1.1 LO Vector Builder Fenêtre glissante (SW LO) Cp1.3 LEP Extractor Extraction des préférences pédagogiques

(Sessionizing) component 2

Learners’ Models Builder

Cp2.1 LK Builder Cp2.2 LEP Builder Cp2.3 Groups Builder

(Indexing | Retrieving) component 3

Content Model Builder

Cp3.1 LO converter Cp3.2 Terms Extractor Cp3.3 Indexer

Recommendation Engine

component 4

Cp4.1 Collaborative Recommender Cp4.2 Content Recommender Cp4.3 Hybridizer Input Learner (Active Session) LO

component 5 HARSYPEL Configuration Module

H A R S Y P E L

Logs + DB Usage data Learning Styles Models

HARSYPEL Architecture

ICALT 2013, Beijing, China 15/07/2013

18

Conclusions and Future Work

 Developed an approach for automatically modeling learners (groups) within LMS taking into account their educational preferences and learning styles  Proposed approach falls within the scope of building an educational automatic hybrid recommender system providing suitable recommendations to learners for personalized technology enhanced learning ;

Future work:

 Evaluation of the recommender system

ICALT 2013, Beijing, China 15/07/2013