4.2 Our approach
4.2.2 Estimating comparison indicators
Chapter 4: Recommendation of Advanced Learners 67
68 Chapter 4: Recommendation of Advanced Learners
Learning processes
& learning activities
……
……
……
Estimate comparison
indicators
Input Ex:
CoW32=0.4524 AoR32=0.7500 C3=4.2
……
Output Ex:
P1ia=0.6581 P2ia=0.7412 P3ia=0.4533
Definition & Calculation
Web Page
Abstract
Remark
Comment
Comment
å å
Î
= Î
ai ai
WP
k ki
WP
k ka
ia CoW
CoW p1'
å
Î=
U i
ia ia
ia p p
p1 1' / 1' 1.Indicator of Comprehension of WP
å å
Î
= Î
ai ai
WP
k ki
WP
k ka
ia AoR
AoR p'2
å
Î=
U i
ia ia
ia p p
p 2'
' 2
2 /
2. Indicator of Adequacy of Remark
åÎ
=
U
i ia
ia
ia p p
p3 3' / '3
å å å å
Î Î
Î Î
=
ai ai
ai ai
WP
k lNote kli
WP
k lNote
kla
ia AoC
AoC p3'
3. Indicator of Agreement of Comment
ia ia
ia
ia p p p
p =a´ 1 +b´ 2 +g´ 3 Combination: linear regression α, β, γ: estimated
User1 User2User3…. Userm
User1 0 0 0.2334 … 0
User2 0.7531 0 0.4553 0
User3 0.2469 0.5 0 … 0.1725
…
Userm 0 0.5 0 … 0
e.g. entry1,3:
The indicator that user3 is more advanced (knowledgeable) than user1 is 0.2334
Fig.4.5 Example of estimating comparison indicators
Comparison indicators are utilized to compare the knowledge level of two learners. Comparison indicators between a pair of learners represent the amount of relationship between them. Indicator of comprehension of WP is based on analysis of web pages and abstracts which are post by both learners. Indicator of adequacy of remark is based on remarks and comments.
Indicator of agreement of comment is based on comments.
whose comments are better than the others’ comments. Fig.4.5 is the example of estimating comparison indicators.
Comparison indicators between a pair of learners represent the amount of relationship between them.
By using extracted interaction indicators described in section 4.2.1, we consider three types of comparison indicators to compare the knowledge level between two learners.
1) The comparison indicator of CoW (Comprehension of web page)
After searching, gathering and aggregating, the learner wrote the abstract. The comparison indicators of CoW are utilized to describe a knowledgeable degree of effective learners among group learner.
By using the analysis of interaction indicator CoW, we can obtain the effective learners’ comparison.
Chapter 4: Recommendation of Advanced Learners 69
å å
Î
= Î
ai ai
WP k
ki WP k
ka
ia CoW
CoW
p1' (4.4)
Then, we use
å
Î
=
U i
ia ia
ia p p
p1 1' / 1' to normalize p1ia' .
Where,
p
1ia is the comparison indicator of learneru
a being more advanced (knowledgeable) than learneru
i based on the interaction indicator comprehension of web page. WPai is the web pages that were learned by both learneru
a and learneru
i.2) The comparison indicator of AoR (Adequacy of remark)
After learners sought the useful or interesting topics on the web, they wrote their remarks about it.
Remarks are produced by the learners. The comparison indicators of AoR are utilized to evaluate a knowledgeable degree of effective creators among group learner. Likewise, we can obtain the effective creators’ comparisons by analyzing the interaction indicator AoR.
å å
Î
= Î
ai ai
WP k
ki WP k
ka
ia AoR
AoR
p'2 (4.5)
Then, we use
å
Î
=
U i
ia ia
ia p p
p2 2' / 2' to normalize p2ia' .
Where, p2ia is the comparison indicator of learner
u
a being more advanced (knowledgeable) than learneru
i based on the interaction indicator adequacy of remark. WPai is the web pages that were learned by both learneru
a and learneru
i.3) The comparison indicator of AoC (Agreement of comment)
70 Chapter 4: Recommendation of Advanced Learners
In comment space, learners can communicate with each other, improvement knowledge by posting self opinions or ideas, generate production together. The comparison indicators of AoC are utilized to calculate a knowledgeable degree of effective collaborators among group learner. We can obtain the effective collaborators’ comparisons through the analysis of factor AoC.
å å å å
Î Î
Î Î
=
ai ai
ai ai
WP
k l Note
kli WP
k l Note
kla
ia AoC
AoC
p3' (4.6)
Then, we use
å
Î
=
U i
ia ia
ia p p
p3 3' / 3' to normalize p3ia' .
Where, p3ia is the comparison indicator of learner
u
a being more advanced (knowledgeable) than learneru
i based on the interaction indicator agreement of comment. WPai are the web pages that were learned by both learneru
a and learneru
i. Noteai are the notes that were wrote comments by both learneru
a and learneru
i.4) Combining comparison indicators
In order to combine the three indicators described above to obtain final result, there are several methods. For example we can use Bayesian or probabilistic networks. However, Bayesian network requires intensive data collection. In addition, computation of prior probabilities, conditional and joint probabilities and relationships among various learner interactions in real time is intensive especially if the network is large and updates are frequent. The simplest way to combine multiple classifiers is by voting, which corresponds to taking a linear combination of the indicators. This is also known as ensembles and linear opinion pools. Here we adopt this simplest way to our approach.
) 1 , 0 ( , ,
1
3 2
1
Î
= + +
´ +
´ +
´
=
g b a
g b a
g b
a ia ia ia
ia p p p
p
(4.7)
Chapter 4: Recommendation of Advanced Learners 71
User1 User2 User3 …. Userm User1
User2 User3
… Userm
……
U1 U2
Um ……
……
…… ……
U1 U2
Um
……
U1 U2
Um
……
User1 User2 User3 …. Userm User1
User2 User3
… Userm
User1 User2 User3 …. Userm User1 0.0167 0.6945 0.2389 … 0.0167 User2 0.1667 0.1667 0.1667 0.1667 User3 0.2268 0.4265 0.0167 … 0.4474
…
Userm0.0167 0.0167 0.1720 … 0.0167
Learning resource
Transition probability matrix indicates changes in knowledge-level comparison of learners over time (with regards to learning process)
Based on comparison indicators
Time = 1 Time = n (future)
Entry 3,1:
The probability that user3 is more advanced (knowledgeable) than user1 is 0.2268
learning process
Time = 2
Learners’ comparison
÷÷
÷÷
÷÷
÷÷
ø ö
çç çç çç çç
è æ
=
0 0 8275 . 0 1725 . 0 0 0
6175 . 0 0 3825 . 0 0 0 0
4785 . 0 5215 . 0 0 0 0 0
0 3113 . 0 0 0 4553 . 0 2334 . 0
0 0 0 0 0 0
0 0 0 2469 . 0 7531 . 0 0 P
÷÷
÷÷
÷÷
÷÷
ø ö
çç çç çç çç
è æ
= -+
=
0167 . 0 0167 . 0 7615 . 0 1720 . 0 0167 . 0 0167 . 0
5725 . 0 0167 . 0 3610 . 0 0167 . 0 0167 . 0 0167 . 0
4474 . 0 4861 . 0 0167 . 0 0167 . 0 0167 . 0 0167 . 0
0167 . 0 2967 . 0 0167 . 0 0167 . 0 4265 . 0 2268 . 0
1667 . 0 1667 . 0 1667 . 0 1667 . 0 1667 . 0 1667 . 0
0167 . 0 0167 . 0 0167 . 0 2389 . 0 6945 . 0 0167 . 0
) 1
( M
P ee P
T
a
& a
&
&
÷÷
÷÷
÷÷
÷÷
ø ö
çç çç çç çç
è æ
= ïî ïí
ì >
= å
0 0 8275 . 0 1725 . 0 0 0
6175 . 0 0 3825 . 0 0 0 0
4785 . 0 5215 . 0 0 0 0 0
0 3113 . 0 0 0 4553 . 0 2334 . 0
1667 . 0 1667 . 0 1667 . 0 1667 . 0 1667 . 0 1667 . 0
0 0 0 2469 . 0 7531 . 0 0
1 .
, 0 otherwise M
P if P
P l
il il
state1 state2 staten
Fig.4.6 Example of estimating learners’ comparison
The learning processes change with regards to time and learning steps. For this reason, we identify the relationship of learners by using Markov Chain Model which includes the learning processes of every learner. Transition probability matrix indicates changes in knowledge level comparison of learners over time (with regards to learning process). P is the translation matrix of combination of three comparison indicators. Pis the revised transition probability matrix. P&&& is the primitive stochastic matrix.
a,b,g are parameter representing the impact of indicator of CoW, the impact of indicator of AoR and the impact of indicator of AoC, repressively.