Personalized elearning system using Item Response Theory

1
Personalized e-learning system using Item
Response Theory

• Authors Chih-Ming Chen, Hahn-Ming Lee
  and Ya-Hui Chen
• Source Computer Education,
• Accepted 16 January 2004
• Speaker Chiang, Hui-Chun ( ???)
• Date 2004/09/18

2
Outline

• Introduction
• System designing and implementation
• Experiments and results
• Conclusions
• Comments
• Appendix

3
Outline

• Introduction
• System designing and implementation
• Experiments and results
• Conclusions
• Comments
• Appendix

4
Introduction

• The shortcomings of most personalization systems
  
• Emphasizing on learner preferences, interests,
  browsing behaviors
• ?Overload of cognitive
• The rapid growth of information on the Web
• ?Overload of information

5
Introduction

• Improving above drawbacks
• ?Personalized e-learning system using Item
  Response Theory (PEL-IRT)
• Considering different levels of learners
  knowledge
• ?Dynamically estimating learners ability based
  on the maximum likelihood estimation (MLE)
• Considering different course material difficulty
• ?Collaborative voting approach for adjusting the
  difficulty

6
Outline

• Introduction
• System designing and implementation
• Experiments and results
• Conclusions
• Comments
• Appendix

7
System designing and implementation

• System architecture

Courses Database
10
User Account Database
Personalized Agent (back-end)
12
2
Courses Recommend- ation Agent
Front-end
User Profile Database
Interface Agent
1
4
5
/ 15
13
14

• Very Hard
• Hard
• Middle
• Easy
• Very Easy

3
11
9
Feedback Agent
7
6
Yes or No
8
8
System designing and implementation

• The learning process of PEL-IRT

9
Outline

• Introduction
• System designing and implementation
• Experiments and results
• Conclusions
• Comments
• Appendix

10
Experiments and results

• Environment Win 2000
• Web server IIS 5.0
• The front-end script language PHP 4.3, MySQL
  server
• Web-site http//203.64.142.234
• The number of learners logged in the system 210
• The number of user profile database 2525
  records
• Object learners for Masters degrees and taking
  neural networks course

11
Experiments and results

• The entire layout of user interface

12
Experiments and results

• An example of course material recommendation
  based on learner ability

13
Experiments and results

• Adjusting the difficulty of course material

14
Experiments and results

• The adaptation of learner ability

15
Experiments and results

• The relationship between learner As ability and
  the difficulty parameter of the recommended
  course material

16
Experiments and results

• Evaluating degree of satisfaction

17
Experiments and results

• Evaluating the learners responses

18
Outline

• Introduction
• System designing and implementation
• Experiments and results
• Conclusions
• Comments
• Appendix

19
Conclusion

• PEL-IRT
• According to course materials visited by learners
  and their responses
• Based on learner abilities, providing
  personalized recommendations course
• Automatically adjusted course difficulty by
  collaborative voting
• Learners only need to reply to two simple
  questionnaires for personalized services

20
Outline

• Introduction
• System designing and implementation
• Experiments and results
• Conclusions
• Comments
• Appendix

21
Comments

• Providing a conscientious method to write papers
• Instead of the questionnaires, using the pretest
  and posttest to get the beginners ability, the
  experienced users ability and the difficulty of
  the course materials

22

• Appendix

23
Operation flowchart of feedback agent

• Three main operations

(1)
(2)
(3)
MLE
Collaborative voting
24
Tuning difficulty parameters of course materials

• Definition 3.1 Difficulty levels of course
  material
• D D1 , D2 , D3 , D4 , D5
• a set of course material difficulty levels
• where D1 -2 very easy
• D2 -1 easy
• D3 0 moderate
• D4 1 hard
• D5 2 very hard

25
Tuning difficulty parameters of course materials

• Definition 3.2 Average difficulty of the jth
  course material based on learner collaboration
  voting
• b j ( voting ) the average difficulty of jth
  course material after learners give collaborative
  voting
• n i j the number of learners that give feedback
  responses belonging to the i t h difficulty level
  for the j t h course material
• N i j the total number of learners that rate
  the j t h course material

26
Tuning difficulty parameters of course materials

• Definition 3.3 The tuned difficulty of course
  material
• bj ( tuned ) w x bj ( initial ) ( 1 w ) x
  bj ( voting )
• bj ( tuned ) the tuned difficulty of the j th
  course material based on learner collaborative
  voting
• bj ( initial ) the initial difficulty of the j
  th course material given by course experts
• w an adjustable weight

27
Estimation of learner abilities

• Estimating learners ability by MLE
• U j 1 or 0 can or cannot completely
  understand the course
• P j(?) , Q j(?) the probability that learners
  can completely understand the jth course material
  at a level below their ability level ?

28
Operation flowchart of courses recommendation
agent
Information function ( Hambleton 1991)
? learner new ability Ij (?) P( a correct
response to the jth course material for learners
with ability ?)
29
Maximum Likelihood Estimation (MLE)

• To determine the parameters that maximize the
  probability (likelihood) of the sample data
• X1,X2,,Xn are random variables ,
• ? is unknown constant parameters which need to be
  estimated
• Likelihood function L(?)g(X1,X2,,Xn ? ),
  while?? ,let L(?) maximum, then ? is the MLE
  of ?