Applying Fuzzy Theory in Intelligent Web Systems

1
Applying Fuzzy Theory in Intelligent Web
Systems

• Chih-Ming Chen (???)
• ?????????
• ??????/???????
• ????????
• ???????????????

?????? ?????????
???????????? ???????????
??????????? ????????????
????????? PDA?????????????
?????????
2
Outline

• The Used Fuzzy Concept
• Fuzzy Inference
• Fuzzy Association Rule
• Neuro-fuzzy network
• Current Researches Relating to Fuzzy Theory in
  Intelligent Web-based Learning Systems
• Personalized Mobile Learning System based on
  Fuzzy Item Response Theory for Promoting English
  Vocabulary and Reading Abilities
• Mining Formative Evaluation Fuzzy Rules Using
  Learning Portfolios for Web-based Learning
  Systems
• The Other Intelligent Systems
• News Archive and Data Mining Agent System
• Chinese Word Segmentation System with Intelligent
  New Word Extension
• Q A Time

3
Fuzzy Inference (1/3)
???????? (??) ???? (??) ??? (??)
???? Fuzzy???? (??) ????? (??) ????????? (??)
??????
4
Fuzzy Inference (2/3)
(facts) X is (rule) if X is A then Y is
B ???????? (result) Y is
Mamdani ?
A
B
1
1
0
0
5
Fuzzy-logic inference system (3/3)
6
Fuzzy Association Rule (1/8)

• The KDD (Knowledge discovery in database) process
  generally consists of the following three phase
• (1) Pre-processing.
• (2) Data-mining.
• (3) Post-processing.
• Fuzzy Transaction Data-mining Algorithm (FTDA)
• This is integrates fuzzy-set concepts with the
  apriori algorithm and uses the result to find
  interesting item-sets and fuzzy association
  rules.

7
Fuzzy Association Rule (2/8)
An example for the detailed process of Apriori
algorithm
8
Fuzzy Association Rule (3/8)
An example for mining fuzzy association rule
Case OOP DB ST DS MIS
1 86 77 86 71 68
2 61 87 89 77 80
3 84 89 86 79 89
4 73 86 79 84 62
5 70 85 87 72 79
6 65 67 86 61 87
7 71 87 75 71 80
8 86 69 64 84 88
9 75 65 86 86 79
10 83 68 65 85 89
9
Fuzzy Association Rule (4/8)
The defined fuzzy membership function for mining
fuzzy association rule
10
Fuzzy Association Rule (5/8)
11
Fuzzy Association Rule (6/8)
Case OOP.Middle DB.High OOP.MiddlenDB.High
1 0.0 0.0 0.0
2 0.0 0.8 0.0
3 0.1 0.9 0.1
4 1.0 0.7 0.7
5 0.7 0.6 0.6
6 0.2 0.0 0.0
7 0.8 0.8 0.8
8 0.0 0.0 0.0
9 0.8 0.0 0.0
10 0.2 0.0 0.0
Count 3.8 3.8 2.2
12
Fuzzy Association Rule (7/8)
13
Fuzzy Association Rule (8/8)
Itemset Itemset Count
OOP.Middle DB.High 2.2
OOP.Middle ST.High 1.8
OOP.Middle DS.Middle 1.7
OOP.Middle MIS.High 0.9
DB.High ST.High 2.2
DB.High DS.Middle 2.7
DB.High MIS.High 1.4
ST.High DS.Middle 2.8
ST.High MIS.High 1.8
DS.Middle MIS.High 1.1
14
Fuzzy-neuro Network
Four-layer learning architecture of the
neuro-fuzzy networks
15
Current Researches Relating to Fuzzy Theory in
Intelligent Web Systems

• Personalized Mobile Learning System based on
  Fuzzy Item Response Theory for Promoting English
  Vocabulary and Reading Abilities

16
Introduction

• The learning form is dramatically changing
• E-learning (Electronic-learning)
• M-learning (Mobile-learning)
• U-learning (Ubiquitous-learning)
• Mobile learning
• an effective form of flexible learning
• utilizing spare time for learning
• learning takes place anytime, anywhere

17
Introduction

• English is an international language.
• English as Second Language, ESL
• English as Foreign Language, EFL
• How to learn English well ?
• assistant tools
• good material
• little and often

18
Purpose of the Study

• Considering the advantages of the mobile learning
• Breaking the limitations of time and space
• Utilizing spare time for learning
• A personalized intelligent m-learning system
  (PIMS) for supporting effective English learning

19
System Design

• A Personalized Intelligent M-learning System
  (PIMS) includes
• The remote courseware server
• The client mobile learning system
• The feature of PIMS
• Portable
• Personalization
• Intelligent tutoring system

20
PIMS System Architecture
21
System Architecture of Personalized Vocabulary
Learning System
22
The Remote Courseware Server

• English news crawler agent
• automatically retrieve English News from the
  Internet
• Difficulty assessment agent of English news
• automatically measuring the difficulty parameters
  of English news articles
• Courseware management agent
• online courseware management

23
The Client Mobile Learning System

• Learning interface agent
• providing a flexible learning interface
• Feedback agent
• collecting learner explicit feedback information
• Personalized courseware recommendation agent
• recommending a personalized courseware
• evaluating learners reading ability

24
The Client Mobile Learning System

• Personalized vocabulary recommendation agent
• enhancing learner vocabulary ability
• discovering the new vocabularies to individual
  learners

25
The Learning Procedure of the Client Mobile
Learning System
26
English E-News Archive

• FTV English e-news
• Metadata extraction mechanism
• English and Chinese news titles
• URL address
• Date
• News body

27
The detailed procedures of English news archive
28
Measuring Difficulty of English News Article

• Readability
• Flesch reading ease formula, 1948
• The drawback of Flesch formula
• No consideration of the readers vocabulary
  ability
• Modified Flesch reading ease formula
• Flesch RE
• Proposed fuzzy difficulty parameter

29
Flesch Reading Ease Formula

• Fleschs reading ease formula can be formulated
  as follows
• RE represents the reading ease value ,range from
  0 (difficulty)100(easy)
• ASL is the average sentence length
• ASW is the average number of syllables per word

30
The Proposed Scheme for Evaluating Difficulty of
English News Article

1. Computing the Percentages of Vocabulary
2. Determining Fuzzy Membership Functions by the
   K-means Clustering Algorithm
3. Designing Fuzzy Rule Base
4. Fuzzy Inference

31
The determined fuzzy membership functions for the
percentage of occurring vocabulary of the
elementary level
32
The defined membership functions for the
difficulty of English News
33
The fuzzy rule base designed by English course
experts for inferring difficulty of English news
article
34
Defuzzification
The center of gravity (COG)
35
An Example for Inferring the Difficulty of
English News Article
36
Computing the value of English news by Fleschs
reading ease formula
Step1
Normalizing the RE value
Step2
37
Computing the difficulty of English news by fuzzy
inference
Step3
The triggered consequent parts of output variable
for defuzzification
38
Determining the final difficulty of English news
article by integrating the normalized and the
inferred difficulty values under the adjustable
weight is set to 0.5
Step4
39
Personalized English News Recommendation

• Item Response Theory (IRT)

40
Personalized English News Recommendation

• Evaluating English reading ability
• the Bayesian estimation approaches is applied in
  this study

41
Personalized English News Recommendation

• Recommending English news
• the maximum information strategy

42
Personalized English News Recommendation

• The drawbacks of item response theory
• Learners response is not usually belonging to
  completely understanding or not understanding
  case for the content of learned courseware
• The traditional item response theory cannot
  estimate learner ability for personalized
  learning services according to learners
  non-crisp responses (i.e. uncertain/fuzzy
  responses)

43
Personalized English News Recommendation

• Fuzzy Item Response Theory (FIRT)

44
Personalized English News Recommendation

• The designed fuzzy rule base for inferring
  learners understanding degree

45
Personalized English Vocabulary Recommendation

• Personalized English Vocabulary Learning System
• Learners vocabulary ability
• Vocabulary difficulty parameter
• Rij the set of the recommended new vocabularies
• Ai the set of vocabularies that the
  corresponding difficulty parameters are higher
  than learners vocabulary ability
• Cj the set of all vocabularies contained in the
  English news article
• Li the set of the acquired vocabularies of the
  learner

46
The Implemented System (1/8)
(b)
(a)
47
The Implemented System (2/8)
(d)
(c)
48
The Implemented System (3/8)
(f)
(e)
49
The Implemented System (4/8)
(h)
(g)
50
The Implemented System (5/8)
(j)
(i)
51
The Implemented System (6/8)
(l)
(k)
52
The Implemented System (7/8)
(n)
(m)
53
The Implemented System (8/8)
(p)
(o)
54
The English Courseware Management System
55
The English Courseware Management System
56
The English Courseware Management System
57
The English Courseware Management System
58
Location-based English Learning System
59
Current Researches Relating to Fuzzy Theory in
Intelligent Web Systems

• Mining Formative Evaluation Fuzzy Rules Using
  Learning Portfolios for Web-based Learning
  Systems

60
Research background

• Learning performance assessment
• To evaluate what learners learnt during the
  learning process.
• The learning performance evaluation instruments
  could be classified as
• the summative evaluation examination
• the formative evaluation learning portfolio

61
Research background (cont.)the summative
evaluation v.s. the formative evaluation.

• The summative evaluation
• The traditional summative evaluation through
  performing examinations or feedback forms to
  evaluate the learning performance for both the
  conventional classroom learning and web-based
  learning.
• To consider final learning outcomes without
  considering the learning process of learners.
• The formative evaluation
• By the formative assessment, teachers feed
  information back to students in ways that enable
  students to learn better, or when students can
  engage in a self-reflective process
• The traditional portfolio v.s. the web-based
  learning portfolio
• Traditional portfolio assessment relies on
  man-made data collection and a writing-centered
  learning process
• The web-based learning portfolio can be
  collected, stored, and managed automatically by
  computers when learners interact with an
  e-learning platform

62
Research background (cont.)

• Data mining
• Data mining is an appropriate method of knowledge
  discovery to excavate the implicit information
  from large repositories.
• The interpretable knowledge could be discovered
  by methods of data mining.
• Grey relational analysis
• Fuzzy association rule
• Fuzzy inference

63
Motivation

• The traditional web-based learning assessment
  only considers final outcomes without considering
  the learning process of learners.
• Web-based learning portfolio could reveal the
  authentic learning process and be recorded
  automatically by computer.
• To perform the learning performance assessment
  using web-based learning portfolio or log data is
  becoming a critical issue in the web-based
  learning field.

64
Contribution

• Teachers could understand the factors that affect
  learning performance in a web-based learning
  environment according the obtained interpretable
  learning performance assessment rules.
• The proposed learning assessment approach can
  correctly evaluate learners learning performance
  according to their learning portfolios.
• For Instructor
• To help teachers to precisely perform the
  formative assessment for individual learner
  utilizing only the learning portfolio in a
  web-based learning environment during learning
  processes.
• For Learner
• The evaluation results of learning are applied to
  help teachers immediately examine learning
  progress of learners and perform interactively
  on-line control learning.

65
System architecture of personalized e-learning
system (PELS)
66
User Interface of personalized e-learning system
(PELS)
67
The proposed learning performance assessment agent

• The gathered learning portfolio on PELS
• Flowchart of the proposed learning performance
  assessment agent
• The used methods

68
The gathered learning portfolio on PELS( 1/3 )
69
The gathered learning portfolio on PELS ( 2/3
)-the learning factor RR-

• The reading rate of course materials is defined
  as the rate of studying course materials in a
  course unit.

A calculating example for the learning factors
of reading rate
70
The gathered learning portfolio on PELS(3/3 )

• The learning portfolio obtained from the
  three-years students of Taipei County Jee May
  Elementary School.
• The total number of learning records is 583.
• Training data 400 learning records
• Test data 183 learning records

71
The flowchart of the learning performance
assessment agent - Scheme 1
72
The flowchart of the learning performance
assessment agent - Scheme 2
73
The Used Methods

• Grey relational analysis for learning factor
  analysis
• Fuzzy association rule mining for learning
• performance assessment rule discovery
• Fuzzifying
• Calculating large itemsets
• Discovering fuzzy rules
• Fuzzy inference
• Fuzzy inferring
• Defuzzification

74
Grey Relational Analysis

• Performing learning factor analysis
• Comparing the grey relational grades between the
  referred sequence Grade(I0 ) and comparative one
  RR,RT.(ii )

Relative learning portfolio
Primitive learning portfolio
75
Fuzzy association rulefuzzifying (1/4)

• To discover fuzzy rules of various grade levels,
  the learning portfolio are partitioned into 5
  non-overlapped groups including Grade.VH,
  Grade.H, Grade.M, Grade.L, and Grade.VL according
  to the membership degrees mapped to each cluster
  center of final testing score Grade.

76
Fuzzy association rulefuzzifying (2/4)

• To classify portfolio based on the learning
  factor Grade by the membership function

77
Fuzzy association rulefuzzifying (3/4)

• The five groups partitioned by the grade level
  for mining the learning performance assessment
  rules and verifying the performance of learning
  assessment

78
Fuzzy association rule fuzzifying (4/4)

• Those quantitative learning record data of the
  portfolio must be transferred into fuzzy learning
  record data for fuzzy association rule mining.
• The clustering centers of five learning factors
  can be appropriately determined by K-means
  clustering algorithm, thus determining the
  membership functions of triangle fuzzy sets.

79
Fuzzy association rulelarge itemset

• The fuzzy rules composed by the large itemset

80
Fuzzy association rulefuzzy rules base (1/2)

• To ensure that the discovered rules are
  interesting and accurate, three conditions that
  include the minimum support , minimum confidence
  and minimum certainty factor are deliberated

81
Fuzzy association rulefuzzy rules base (2/2)

• From the discovered fuzzy rules, we find that
• No rules are discovered in the grade level G.VL
  due to fewer training data

The level of Grade The amount of rules
Grade.VL 0
Grade. L 6
Grade.MID 7
Grade.H 11
Grade.VH 14
The amount of rules in 5 groups
82
Fuzzy inferenceFuzzifying

• The following table is an example of learning
  portfolio.
• To infer the grade of the final test for this
  learner according to the learning factors RR, RT,
  LA, CR,and HL
• The following table presents fuzzy values and
  level of each item transferred by the membership
  function

83
Fuzzy inferenceFuzzy inferring (1/4)

• The following 11 rules are triggered by the
  learning portfolio of the learner with student ID
  11737

84
Fuzzy inferenceFuzzy inferring (2/4)
-Grade.H


85
Fuzzy inferenceFuzzy inferring (3/4)
-Grade.MID
86
Fuzzy inferenceFuzzy inferring (4/4)
-Grade.L
87
Fuzzy inference defuzzification

• Real_grade 86.3043
• Infer_grade 86.3076

88
Fuzzy inference
89
Experiments

• Accuracy rate of the learning performance
  assessment
• 5points
• The score level
• To consider various combinations of learning
  factors
• A, B, C, D, ECR-LA-HL-RT-RR, CR-LA-HL-RT,
  CR-LA-HL, CR-LA, CR
• The infer example of A, C, E

90
Experiments 5 points

• The method of 5 points is used to evaluate the
  accuracy rate of the predicted learning
  performance. That is, if the difference of the
  predicted test score with the actual test score
  is between -5 points to 5 points, then the
  predicted result is served as correct.

83.33
91
Experiments score level

• The score level is judged as Grade.M if the
  actual test score of some learner is 85.16
  because the linguistic term of the Grade.M has
  largest mapped membership degree compared to the
  other linguistic terms of the final grade

66.67
92
various combinations of learning factors the
infer example of A, C, E

• The following chart present the accuracy of the
  infer_grade of 3 itemsets, A?C?E
• ACR, LA, HL, RT, RA
• 33.33
• CCR, LA, HL
• 83.33
• E CR
• 83.33

93
Experiments various combinations of learning
factors

• The predicted accuracy rates of 400 training data
  and 183 testing data under considering various
  combinations of learning factors for the two
  proposed accuracy evaluation methods
• The combination of factors are more relative with
  learning performance, the accuracy is better.

94
Experiments accuracy rate of factor CR

• Comparison results of the predicted accuracy
  rates of various grade levels for 400 training
  data and 183 testing sample under only
  considering the learning factor CR

95
The Other Intelligent Web Systems

• News Archive and Data Mining Agent System
• Chinese Word Segmentation System with Intelligent
  New Word Extension

96
New Archive And Data Mining Agent System
97
New Archive And Data Mining Agent System
98
New Archive And Data Mining Agent System
99
Chinese Word Segmentation System with Intelligent
New Word Extension
100
Chinese Word Segmentation System with Intelligent
New Word Extension
101
Chinese Word Segmentation System with Intelligent
New Word Extension
The designed fuzzy rule base for inferring the
confidence degree of new word
102
Q A TimeThanks for your listening!