A new approach for building student model in an
Adaptive and intelligent Web-Based Educational System
(AIWBES) is introduced. This approach utilizes a hybrid
algorithm based on Fuzzy-ART2 neural network and stochastic
method called Hidden Markov Model (HMM), in order to
evaluate and categorize students’ knowledge status in six levels:
Excellent, very good, good, fair, weak and very weak; depending
on 5 parameters collected through their interactions with the
system. The student model is initialized by presenting a pre-test
form to students and it is updated dynamically according to their
study times and assessment results. Students' knowledge status
are modeled through three phases, initialization, training and
recall phases. In the initialization phase, input vectors are
normalized before they are categorized using unsupervised
algorithm Fuzzy-ART2 in 6 clusters representing 6 knowledge
status. A HMM is created for each cluster and when new
students' parameters are collected, they are introduced to Baum-
Welch re-estimation algorithm to train the 6 HMMs and to
maximize the observed sequence that is associated with a
particular cluster. Forward algorithm evaluates then the
likelihood of this sequence with respect to each of the HMMs and
to determine the maximum value, which represents the actual
knowledge status of the student. Experiment results show that
the proposed approach is capable of categorizing student
parameter vectors to their corresponding cluster with good
accuracies. The result of such classifications would open new
horizons and applications in AIWBES.
Call for Papers - African Journal of Biological Sciences, E-ISSN: 2663-2187, ...
Student Modeling Using NN-HMM for EFL Course
1. Student Modeling Using NN-HMM for EFL Course
International Conference on Information & Communications Technologies from Theory to Applications –ICTTA -08, 06-10 April 2008, Damascus-Syria
Student Modeling Using NN-HMM for
EFL Course
Masun Nabhan Homsi
University of Aleppo
SYRIA
Dr.Rania LUTFI
University of Al-Baath
SYRIA
Prof. Dr Ghias BARAKAT
University of Aleppo
SYRIA
Rosa María Carro Salas
Universidad Autónoma de
Madrid
2. Student Modeling Using NN-HMM for EFL Course
International Conference on Information & Communications Technologies from Theory to Applications –ICTTA -08, 06-10 April 2008, Damascus-Syria
Overview (1)
• Introduction
• Adaptive and Intelligent Web-Based
Educational System (AIWBES)
• What is Student Model (SM)?
• Fuzzy Adaptive Resonance Theory
(Fuzzy-ART2)
• Hidden Markov Model (HMM)
• The new system architecture
3. Student Modeling Using NN-HMM for EFL Course
International Conference on Information & Communications Technologies from Theory to Applications –ICTTA -08, 06-10 April 2008, Damascus-Syria
Overview (2)
• NN-HMM Student model
– Initialization Phase
– Training Phase
– Recall Phase
• System Implementation
• Results
• Conclusions
• Future works
4. Student Modeling Using NN-HMM for EFL Course
International Conference on Information & Communications Technologies from Theory to Applications –ICTTA -08, 06-10 April 2008, Damascus-Syria
Introduction (1)
•Adaptive and Intelligent Web-based educational systems
(AIWBESs) use Artificial Intelligence (AI) Techniques in
order to adapt mainly to student needs for self-study.
•Student Model
It is the core component of many adaptive and intelligent
web-Based Educational Systems (AIWBES). Student model
enables these systems provide individualized course content
and study support to help students with different
background and knowledge status to achieve their learning
objectives.
5. Student Modeling Using NN-HMM for EFL Course
International Conference on Information & Communications Technologies from Theory to Applications –ICTTA -08, 06-10 April 2008, Damascus-Syria
• Unsupervised network capable of retaining plasticity
throughout life, without endangering the stability of
previously learned patterns, it consists of:
• Basic Architecture
– Input Field F0
– Input Field F1
– Output Field F2
– Orienting Subsystem
Fuzzy-ART2 (2)
Input Patterns
Top-Down
LTM
STM F0
STM F2
STM F1
Bottom-Up
LTM
6. Student Modeling Using NN-HMM for EFL Course
International Conference on Information & Communications Technologies from Theory to Applications –ICTTA -08, 06-10 April 2008, Damascus-Syria
• HMM (Hidden Markov Model) is a doubly embedded
stochastic process with a process that is not
observable (hidden process) and can only be
observed through another stochastic process
(observable process) that produces the time set of
observations.
• HMM is initially developed for speech recognition
approach and it is subsequently used in different and
numerous applications such as bioinformatics,
gesture recognition, music classification, machine
translation and others.
Hidden Markov Model (1)
7. Student Modeling Using NN-HMM for EFL Course
International Conference on Information & Communications Technologies from Theory to Applications –ICTTA -08, 06-10 April 2008, Damascus-Syria
• Problem (1): Given =(A, B, ), how P(O|) is computed?,
the probability of occurrence of the observation sequence
O=O1,O2,…,OT.
– Solution: Forward-Backward algorithm.
• Problem (2): Given =(A, B, ), find a state sequence
I=i1,i2,…iT such that the occurrence of the observation
sequence O=O1,O2,…,OT is greater than from any other
state sequence.
– Solution: Viterbi algorithm (Out of the scope of our paper)
• Problem (3): How the model =(A, B, ) is adjusted to
maximize P(O,I|) or P(O|).
– Solution: K-means algorithm maximizes P(O,I|), and Baum-Welch re-
estimation formulas maximize P(O|).
Hidden Markov Model (2)
8. Student Modeling Using NN-HMM for EFL Course
International Conference on Information & Communications Technologies from Theory to Applications –ICTTA -08, 06-10 April 2008, Damascus-Syria
The New System Architecture
The system consists of
five Components:
• Domain model
• Student model
• Pedagogical module
• Interface module
AIWBES
Domain
Model
Pedagogic
al Module
Student
Model
Interface
XML+DTD
NN-HMM
9. Student Modeling Using NN-HMM for EFL Course
International Conference on Information & Communications Technologies from Theory to Applications –ICTTA -08, 06-10 April 2008, Damascus-Syria
Domain Knowledge
• The domain knowledge tree, as far
as the learning Objective , is
displayed in the navigation area of
the user interface
• From that tree the student can
choose a learning Objective
• Each subsection corresponds to a
Educational Unit, which is a php
page. That is, only learning
Objective correspond to
displayable material.
• The Educational Unit of the
selected learning Objective is
currently presented in the content
area.
• Each Educational Unit deals with a
number of concepts.
• Concepts are classified in:
– main concepts
– Prerequisite concepts
– Sub-Concepts
Sub-
Concept
Sub-
Concept
1
Sub-
Concept
Sub-
Concept
N
…
…
…
Learning
Objective
1
Learning
Objective
N
Concept
N
Concept
1
EFL Course
EU
N
EU
1 …
10. Student Modeling Using NN-HMM for EFL Course
International Conference on Information & Communications Technologies from Theory to Applications –ICTTA -08, 06-10 April 2008, Damascus-Syria
Domain Knowledge
" Present Continuous Tense "
Sub-ConceptsPrerequisite ConceptMain Concept
Verbs of Actions
Verbs of state
Simple Present
Tense
Time expressions
To be verb
Gerund
Positive Form
Positive FormNegative Form
Positive Form
Negative Form
Question Form
Question FormShort and Long
answers
11. Student Modeling Using NN-HMM for EFL Course
International Conference on Information & Communications Technologies from Theory to Applications –ICTTA -08, 06-10 April 2008, Damascus-Syria
Pedagogical Module
• R1: The Concept is chosen if all its prerequisite concepts are
learned.
• R2: The concept is chosen once the student's knowledge status
for the concept is at least "very good".
• R3: A Glossary of sub-concept is suggested to the student,
when his/her knowledge status for a concept is less than "rather
good".
• R4: The concept is not chosen if the student's knowledge status
is less or equal to weak.
• R5: A concept is shown in green color if the student's
knowledge status is fair.
• R6: A concept is shown in blue color if the student's knowledge
status is good.
• R7: A concept is shown in violate color if the student's
knowledge status is very good.
• R8: A concept is shown in black color if the student's knowledge
status is excellent.
12. Student Modeling Using NN-HMM for EFL Course
International Conference on Information & Communications Technologies from Theory to Applications –ICTTA -08, 06-10 April 2008, Damascus-Syria
Student Model
• Static part
It keeps student's personal and educational
information.
• Dynamic part
The main role of this part is to maintain the
record of the student’s understanding as the
course progresses, and it does so, on the
basis of student responses.
13. Student Modeling Using NN-HMM for EFL Course
International Conference on Information & Communications Technologies from Theory to Applications –ICTTA -08, 06-10 April 2008, Damascus-Syria
Interface Module
• Teacher Interface
• Student Interface
– Index area
– Navigation area
– Post-Test area
– Glossary area
Index Area
Post-test
area
Glossary
Area
Navigation
Area
Student's interface
14. Student Modeling Using NN-HMM for EFL Course
International Conference on Information & Communications Technologies from Theory to Applications –ICTTA -08, 06-10 April 2008, Damascus-Syria
NN-HMM Student Model
• Student model is built using three phases :
Initialization
Phase
Training
Phase
Recall
Phase
15. Student Modeling Using NN-HMM for EFL Course
International Conference on Information & Communications Technologies from Theory to Applications –ICTTA -08, 06-10 April 2008, Damascus-Syria
Initialization Phase
- Input Normalization, Input Clustering, Input
symbolization, Initialize HMMs
,A,B
,A,B
,A,B
,A,B
,A,B
,A,B
Clusters
Cluste
ring
Algori
thm
(Fuzzy
-
ART2)
Very Week
Excellent
Very good
Good
Rather good
Week
Norm
alize
Input
vector
s
Inputs
vectors
represent
students'
parameters
for each
concept:
NCA, NICA,
TSSQ, TSR,
NAAQ
Calculate
HMM1
HMM2
HMM3
HMM4
HMM5
HMM6
Create a Hmm
for each cluster
16. Student Modeling Using NN-HMM for EFL Course
International Conference on Information & Communications Technologies from Theory to Applications –ICTTA -08, 06-10 April 2008, Damascus-Syria
Training Phase
1. Baum-Welch algorithm is used to train 1,
2, … k by adjusting their parameters so
as to increase likelihood functions
P(O|1), P(O|2), … , P(O|k) until a
maximum value is reached for each
and to get new models.
.
17. Student Modeling Using NN-HMM for EFL Course
International Conference on Information & Communications Technologies from Theory to Applications –ICTTA -08, 06-10 April 2008, Damascus-Syria
Recall Phase
Initial HMMs
(1,2,3,4,5,6)
New
observation
sequence for a
concept
Baum-
Welch
Algorit
hm for
HMMs
(1,2,3,
4,5,6)
New HMM1
New HMM2
New HMM3
New HMM4
New HMM5
New HMM6
Forward
algorithm
Forward
algorithm
Forward
algorithm
Forward
algorithm
Forward
algorithm
Forward
algorithm
Get maximum
Probability to
determine
knowledge
status of the
student related
to each concept
18. Student Modeling Using NN-HMM for EFL Course
International Conference on Information & Communications Technologies from Theory to Applications –ICTTA -08, 06-10 April 2008, Damascus-Syria
Implementation Issues
Student model is constructed through two stages:
• Initializing Stage: The student model is
initialized when a student logs in the course for
the first time, the system sends a form to be
filled with his/her personal and educational
information. A pre-test session is then
presented to evaluate his/her knowledge status
regarding to all learning objectives, concepts
and sub-concepts.
• Updating Stage : The student model is updated
each time the student interacts with the course
to determine his/her new knowledge status.
19. Student Modeling Using NN-HMM for EFL Course
International Conference on Information & Communications Technologies from Theory to Applications –ICTTA -08, 06-10 April 2008, Damascus-Syria
Implementation Issues (cont.)
A relational database is built with two tables;
• student's table
• Student's parameters table
NAAQTSRTSSQNICANCA#Student
910546431
1
19164232
10439713
3179151904
2 0264495635
…………………
•NCA: The number of correct answers.
•NICA: The number of incorrect answers.
•TSSQ: The time spent to solve a
question.
•TSR: The time spent to reading or
interacting with a specific concept
•NAAQ: The number of attempts to
answer a question.
20. Student Modeling Using NN-HMM for EFL Course
International Conference on Information & Communications Technologies from Theory to Applications –ICTTA -08, 06-10 April 2008, Damascus-Syria
Implementation Issues (cont.)
• A new tool is also built using PHP language to
simulate Fuzzy-ART2 and HMM functions.
Training
Form
Recall
Form
21. Student Modeling Using NN-HMM for EFL Course
International Conference on Information & Communications Technologies from Theory to Applications –ICTTA -08, 06-10 April 2008, Damascus-Syria
Results
Initialization Phase
NAAQTSRTSSQNICANCA#
0.0157340.0174830.9545450.0069930.0052451
0.0062110.5652170.3975160.0124220.0186342
0.0892860.0267860.84821400.0357143
0.0087720.5233920.441520.02631604
00.343750.6445310.0078130.0039065
………………
NAAQTSRTSSQNICANCA#Student
910546431
1
19164232
10439713
3179151904
2 0264495635
…………………
Input Normalization
n
j
j
i
i
x
x
x
0
22. Student Modeling Using NN-HMM for EFL Course
International Conference on Information & Communications Technologies from Theory to Applications –ICTTA -08, 06-10 April 2008, Damascus-Syria
Results
Initialization Phase
Input Clustering
23. Student Modeling Using NN-HMM for EFL Course
International Conference on Information & Communications Technologies from Theory to Applications –ICTTA -08, 06-10 April 2008, Damascus-Syria
Results
Initialization Phase
NAAQTSRTSSQNICANCA#
0.0157340.0174830.9545450.0069930.0052451
0.0062110.5652170.3975160.0124220.0186342
0.0892860.0267860.84821400.0357143
0.0087720.5233920.441520.02631604
00.343750.6445310.0078130.0039065
………………
Input Symbolization
No. of cluster
Observation
Sequence
#
02 1 5 0 01
20 0 3 3 12
50 4 5 0 03
44 0 3 3 14
12 1 3 3 45
.........
24. Student Modeling Using NN-HMM for EFL Course
International Conference on Information & Communications Technologies from Theory to Applications –ICTTA -08, 06-10 April 2008, Damascus-Syria
Results
Training and Recall Phases
65432
No. of States
(N)
26%20%20%23%23%percentage
No. of
cluster
Observation
Sequence
#
02 1 1 0 0 11
22 1 0 2 0 02
53 0 0 0 1 13
41 1 0 2 1 04
10 1 1 2 1 05
.........
A second Symbolization process results
No. of
cluster
Observation
Sequence
#
02 1 5 0 01
20 0 3 3 12
50 4 5 0 03
44 0 3 3 14
12 1 3 3 45
.........
25. Student Modeling Using NN-HMM for EFL Course
International Conference on Information & Communications Technologies from Theory to Applications –ICTTA -08, 06-10 April 2008, Damascus-Syria
Results
Training and Recall Phases
• the percentage of recall performance is
affected by two factors:
– The number of hidden states of HMMs, e.g. No. 3 is
considered the optimal to get the highest
percentage 70%.
– The length of the input vector (observation
sequence) which is now 6.
65432
No. of States
(N)
66%66%66%70%23%
percentage of
Recall
performanc
e
26. Student Modeling Using NN-HMM for EFL Course
International Conference on Information & Communications Technologies from Theory to Applications –ICTTA -08, 06-10 April 2008, Damascus-Syria
Conclusions
• This paper introduces Fuzzy-ART2-HMM that
model students' knowledge for better learning
and inference in AIWBES for teaching EFL.
• This paper makes four main contributions.
• It demonstrates how a hybrid architecture can be
applied to model student's knowledge in
AIWBES.
• It demonstrates a new application of HMM in the
field of education.
• The output of the research may be used as a
benchmark for educationalists.
• The result of this paper can open new horizons
in building others student models in AIWBES
27. Student Modeling Using NN-HMM for EFL Course
International Conference on Information & Communications Technologies from Theory to Applications –ICTTA -08, 06-10 April 2008, Damascus-Syria
Future Work
• Evaluate the course in real educational
situations
• Make several comparisons of results
among variations of student models using
Fuzzy-ART2, HMM or NN-HMM
28. Student Modeling Using NN-HMM for EFL Course
International Conference on Information & Communications Technologies from Theory to Applications –ICTTA -08, 06-10 April 2008, Damascus-Syria
-University of Aleppo
ِ◌
- Al-Baath University
-Universidad Autónoma de Madrid