1. KNOWLEDGE MODELING FROM
ACTIVITY TRACES WORKSHOP
9th Joint Europpean TEL Summer School 2013
Ben-Manson TOUSSAINT Antigoni PARMAXI
Ricardo KAWASE Vanda LUENGO
2. QUESTIONS TO BE ANSWERED
What is Educational Data Mining?
Related techniques, methods and tools?
How EDM helps in education?
28/05/2013
2
3. EDUCATIONAL DATA MINING
Application of Data Mining methods
and techniques on top of data
coming from educational
environments.
In simpler words: it is all about identifying significant patterns in large
educational datasets.
28/05/2013
3
4. QUID SIGNIFICANT PATTERNS?
Co-occurrences: association rule
mining, sequence mining.
Actions A is always followed by decision D;
people that decides D often do B.
Clusters: cluster analysis
Analyzed events have enough distinctive
characteristics to be split into specific sets
Classification: classification rule
discovery.
Elements are identified considering
characteristics of predefined sets
28/05/2013
4
5. QUID SIGNIFICANT PATTERNS?
Co-occurrences: association rule
mining, sequence mining.
Actions A is always followed by decision D;
people that decides D often do B.
Clusters: cluster analysis
Analyzed events have enough distinctive
characteristics to be split into specific sets
Classification: classification rule
discovery.
Elements are identified considering
characteristics of predefined sets
COUNTING
28/05/2013
5
6. EDM CLASSES OF METHODS
Prediction
Clustering
Relationship Mining
Discovery with Models
Distillation of Data For Human Judgment
28/05/2013
6
7. EDM CLASSES OF METHODS
Outlier Detection
Social Network Analysis
Process Mining
Text Mining
Knowledge Tracing
Non-negative Matrix Factorization
28/05/2013
7
8. WHY « ACTIVITY TRACES »?
Activity traces are logs of possibly ALL
the interactions with a learning system.
Everything is/can be recorded: temporal
information, actions, learning context,
task context, etc.
Detailed
Multi
sourced
Diverse
formats
Diverse
structures
Fine
grained
28/05/2013
8
9. WHY « ACTIVITY TRACES »?
Activity traces are logs of possibly ALL
the interactions with a learning system.
Everything is/can be recorded: temporal
information, actions, learning context,
task context, etc.
Better than simply judge
performance
GOOD
BAD
15/20
A+
28/05/2013
9
10. TELEOS: Technology Enhanced Learning
Environment for Orthopedic Surgery
EXAMPLE OF ACTIVITY TRACES
28/05/2013
10
12. Definitely helpful considering the amount of data
Can find information that you were not aware of
Can identify outliers
Can identify slight differences between patterns
28/05/2013
12
15. WELL! WHAT’S THE POINT
How can EDM concretely help in education
28/05/2013
15
16. 1.Improvement
of student
models
Student’s …
Current
knowledge
…Motivation
…Meta-
cognition
…Behavior
2. Discovery or
improvement of
models of a
domain’s knowledge
structure
Interrelationships
of knowledge in a
domain
Automatic
discovery
of domain
structures
3. Study of
pedagogical
support
Discovery of
effective type of
pedagogical
support
4. Highlighting
of empirical
evidence
Extension of
educational
theories
Key factors
impacting
learning
Design of
better learning
systems
28/05/2013
16
17. 1.Improvement
of student
models
Student’s …
Current
knowledge
…Motivation
…Meta-
cognition
…Behavior
2. Discovery or
improvement of
models of a
domain’s knowledge
structure
Interrelationships
of knowledge in a
domain
Automatic
discovery
of domain
structures
3. Study of
pedagogical
support
Discovery of
effective type of
pedagogical
support
4. Highlighting
of empirical
evidence
Extension of
educational
theories
Key factors
impacting
learning
Developing methods for
exploring educational data
and better understand
students, and the settings
which they learn in
www.educationaldatamining.org
Design of
better learning
systems
28/05/2013
17
18. Right dataset
Long dataset
Wide dataset
Detailed
dataset
[Educational] data mining tools and
techniques are generally appropriate on:
28/05/2013
18
22. What kind of educational data do you deal with?
Source.- social network, web learning environment, MOOCS Intelligent Tutoring
System, etc.
Volume.- huge, wide…
Type.- numerical, textual, alphanumerical
Time information.- granularity, no time traces, etc.
Multi-flowed.- are traces for a same activity split into different flows?
Heterogeneity: structure, formats, temporal granularity...
28/05/2013
22
24. What application do you target? Why?
Improve student models
Discover/improve models of domain’s knowledge structure
Assess pedagogical supports
Identify empirical evidence
What features?
Classes of methods: ____________________________________
Preprocessing tasks: ____________________________________
Visualization tasks: _____________________________________
Mining features: ________________________________________
What results are you expecting? For what purposes?
(in a few words)
28/05/2013
24