DiscoveringCollaborative Patterns
in eLearning fromMeta-code Subsequence
Name: Yip Chi KinDate: 15-01-2005
Motivation
․Collaborative eLearning․Machine Learning Assess․Temporal Data Mining․Generalization Usage
eLearning․Learning Style & Model․Collaborative activity․Courseware․Technology․Assessment
Temporal Data
․Asynchronousness․Irregularity․Huge Volume․Streaming Data․Distributed analysis․Heterogeneous data types
SynchronousDataset
Temporal MiningThreshold selection Frequency Analysis Anomaly Detection
Prediction Correlation Regression BenchmarkingCausality Analysis Periodic Pattern MiningSequential Event PatternsTemporal Association FindingClustering and Classification
Bioinformaticsapproach
Tracking Techniques․Full Screen HyperCard
․Click tracking of user․Timeout timestamp․Referrer timestamp․Full-loaded timestamp․Machine hanging․Diminish / Enlarge Platform Window․Delay capture session code
Temporal Timestamp
Event ID User ID Timestamp From Referrer
22345 101 20030620160000 S21 t12
22346 66 20030620160008 PLI ipi
22347 101 20030620160010 T12 cmi
S21, t12, … are HTML page codesipi, cmi, … are communicative sessions
Temporal Grouping
ID User ID Timestamp Group_id
223 101 20030620160000 21
224 66 20030620160008 35
225 101 20030620160010 42
Members could be join to another groupAnytime in the eLearning Platform
Data Cleaning
․Hacking Problem․Graphics Learning․Session Errors․Double Clicks
s = Number of students, where 1 s 163 t = Timestamp c = Pagecode
Duration(Page)= Starting Timestamp(Page) Ending Timestamp(Page) = St Et = Dt = Duration of each page
․Fuzzy rule:
Duration
Browse weight Bc = s ( c Dt )
if Dt 1 second then Dt = 0
s = Number of students, where 1 s 163 t = Timestamp c = Pagecode
Ps = Frequency(Page) of each student
fc = s Ps = Total frequence of page
․Fuzzy rule:if Dt 3 seconds then fc = fc + 1
Frequency
where
Weighting
Bc is Duration of Page Code
fc is Frequency of Page Codec is Linguistic variable of each Events
where
Navigation of Page
c
ccc f
BαX
Navigation Pages Statistics
Raw Browsing data After Logarithm
Normalization
․Unique Interval
Normalization of Page Code ( 0 wn 1 )
where xn is source value and wn is weight value
maxx macima and minx minima for all data
xx
xnn
xw
minmax
min)log(
Events CodingA = ConceptsB = IndividualC = CollaborationD = TechniqueE = eLearningX = Idle Time
Theory, Courseware
Video, Skill
Information Granules
․Linguistic variables․Fuzzy Reasoning․Interval Valued․Super Subsequence
Meta-Code
Code about code
MetaCodeInformation Granulation
Events Code
Interval Valued
Linguistic variables
Fuzzy Rules
Tri-event pattern
Discretization
Timestamp
Time Partitioning
TemporalDatabase Fuzzy Rules
Code Sequence
Temporal Reasoning
Granular Computing
Event Weight
Frequency & Time
Fuzzy Rules
…BC D C B A A C
Linguistic Variables
Contiguous Sequence
Collaborative Window
Communication Window
AD D D A A A AAB B A A A A X AX X X X A A A
BA D B B A D DAD C A A B B B AB D D A A C C
AA A A A A A ACA A A A X A A DA A A D D D D
BC D D D B B BBD D A A A A B AB B A A C C C
DC C D D C C CXC B A A X X A BA A C C B B B ……………
Group Size
Mutual event window size
Individual Personalization Profile
Member #4
Member #2
Member #3
Collaborative LinkIncreasing weights from collaborative links
n is Numbers of LinksW is weight of events
Sc = ncWc
Synchronous Communication
where
Member #1
Member #5
Synchronous Communication
Syne = Sc
e is eventsc is member of group
where
Synchronous Weight
Asynchronous Communication
Asyne = Ac
c
ccc n
WWA
Communication
W is weight of eventsn is Numbers of Asynchronous Communications
where
Asynchronous Weight
Capture Windows
One day period 86400 sec
Effective communications100000 to 1400000 sec
Study period subsequence 6480000 sec
Minimum weighting&
Maximum weighting
Capture direction
Points of Weight = (Asyne + Syne)
Minimum Windowing
0
2
4
6
8
10
12
14
16
18
20
100000 300000 500000 700000 900000 1100000 1300000
G roup 1
G roup 2
G roup 3
G roup 4
Maximum Windowing
0
2
4
6
8
10
12
14
16
18
20
100000 300000 500000 700000 900000 1100000 1300000
G ro u p 1G ro u p 2G ro u p 3G ro u p 4
Result Applications
Coursework (#5) studying period should be more than 6 days
Range of Collaboration benchmark is 77920 to 176497 points of weight
Necessary communication period is 4 days
• Curriculum Planning
• Collaborative Assessment
• Effective Communication
Project Enhancement․Huge Volume Implementation
(Apply special algorithms)
․Rewrite C++ Programs (Generalization Usage)
․Data Organization (One day Members = 86400 163) Vis․ualization of Patterns
MetaCode ModelingTime interval = 1 second , weblog duration = 75 days , Code length of Personalization Profile = n = 75246060 = 6480000
n
… BA A A B D D DBA A B B D C A CA A A C C C A
… A B DCA B D A C A
MetaCode Events SubSequence
MetaCode SubSequence
Tri-event PatternMutual relationships of tri-event pattern in sub-sequence
․Comparison of good/bad tri-event patterns․Frequent sequential pattern finding (tri-event)․Longest common subsequence․Super Subsequence․Sequential events prediction․Sequence reconstruction․Viterbi algorithm (hamming distance + Transformational grammar )
Super Subsequenceset of tri-event: {000, 001, 010, 011, 100, 101, 110, 111}
concatenationSuper Subsequence
: 000 001 010 011 100 101 110 111
Super Subsequence 0001110100
011000
001111 101
110010
100
Further Research
․Automata and Computability․Implementing Algorithms․Fuzzy Linguistic Associative Rules․Fuzzy Reasoning Partitioning․Subsequence Matching․MetaCode Grammar
Conclusions
․Assess Collaboration․Granular Modeling․Generalization Usage