Interpreting Data Mining Results with Linked Data for Learning Analytics:Motivation, Case Study and Directions
Presentation at the LAK 2013 conference - 10-04-2013
Interpreting Data Mining Results with Linked Data for Learning Analytics
1. Interpreting Data Mining Results with
Linked Data for Learning Analytics:
Motivation, Case Study and Directions
Mathieu d’Aquin
Knowledge Media Institute, The Open University
mdaquin.net - @mdaquin
mathieu.daquin@open.ac.uk
Nicolas Jay
Université de Lorraine, LORIA,
nicolas.jay@loria.fr
2. My super naïve view of learning
analytics
Insight!
Tada!
Some kind of data
processing Visualisation
Data (from some
education
related system)
3. But actually…
Insight!
Tadada!
Some kind of data
processing Visualisation
Interpretation
Data (from some
education
related system)
4. Needs more data/information
Insight!
Tadada
Some kind of data dou!
processing Visualisation
Background
Interpretation
Data (from some knowledge
education
related system)
5. The challenge for learning analytics
Most of the time, background knowledge
needs to be in the head of the people looking
at the analytics.
How to find/obtain background information
for interpretation to support him/her
considering that:
– The data we are analysing and insight we are
trying to obtain can cover a wide range of
things, topics, domains, subjects…
– We might not know in advance we background
information is needed for interpretation
Our approach: Integrate linked data
sources at the time of interpretation
6. What’s linked data
See the “Using Linked Data in Learning
Analytics” tutorial yesterday
http://linkedu.eu/event/lak2013-linkeddata-
tutorial/
7. Linked Data
Open University Person: Mathieu
Website
Publication: Pub1
author
workFor
Open University
VLE
Course: M366
offers
KMi Website M366 Course
page
Organisation:
The Open University
Mathieu’s
Homepage availableIn
setBook
Mathieu’s
List of Mathieu’s
Publications Twitter Country: Belgium
Book: Mechatronics
The Web The Web of Linked Data
8. rNews
Music
Ontology Geo
Ontology
SIOC Media
Ontology
Dublin
Core
DBPedia
FOAF
Ontology
DOAP
FMA BIBO
Ontology
LODE
Gene
Ontology
10. Use case: student enrolment data
From the
Open
University’s
Course Profile
Facebook
Application:
Examples:
Who enrolled
to what
Student ID Course Code Status Date
112 dse212 Studying 2007
course at 112 d315 Intend to study 2008
what time 109 a207 Completed 2005
11. Sequence mining
We can represent each student’s trajectory by a
sequence of courses, e.g.
(DD100) (D203, S180) (S283)
Applying sequence mining makes it possible to
find frequent patterns in these sequences, i.e.,
courses often taken together in a certain order.
12. The results
(and again, why they need background knowledge for
interpretation)
Out of 8,806 sequences (students), we obtained
126 different sequential patterns with a support
threshold of 100*
i.e. filtering out patterns included in less than 100 sequences.
Sequential pattern Support
(DD100) (DSE212) 232
Examples: (DSE212) (ED209) (DD303) 150
(B120) (B201) 122
How to know what that means?
We need background information about the
courses (DD100, DSE212, ED209 ,etc.)
14. Making the results linked data
compliant
Use a simple ontology of sequences to represent
the patterns
And use linked data URIs to represent the items,
e.g. DSE212
http://data.open.ac.uk/course/dse212
15.
16. Selecting a dimension in linked data
Propose relations that
apply to the items of
the patterns
Then relations that
apply to the objects of
these relations
Etc.
i.e. follow the links to build a chain of
relationships.
17. Building a hierarchy of patterns
The end-values of the
chain of relations built
out of following links
of linked data form
attributes of the
patterns
Build a lattice
(hierarchy) of
concepts representing
groupings of these
attributes, using
formal concept
analysis
20. Benefits
(see following examples)
Provides an overview of the patterns
obtained along a custom dimension
Helps identifying gaps and issues in
the original data/process
Helps identifying areas in need of
further exploration
Generic: can be straightforwardly
applied to other source data, other
linked data and other mining methods
24. Discussion
Limitations of the approach:
– Requires the results to be linked data and the
items to connect to linked data
– Sources of linked data needs to be available to
support interpretation)
http://data.linkededucation.org/linkedup/catalog
25. Discussion: It’s a loop
Views and
Data selection
dimensions
mining
Background
Interpretation
Data (from some knowledge
education
related system)
26. Conclusion
Linked data can be used to enrich and
bring some meaningful structure to
the patterns from an analytics/mining
process
Introducing linked data not only in
input of the process, but also in
support of more analytical tasks
Promising, considering the growth of
education-related linked data
Should become part of an iterative
process, where patterns and data get
refined through interpretation and the
introduction of background
information from linked data
27. Thank you!
More info at:
http://mdaquin.net @mdaquin
http://linkedup-project.eu
http://linkedup-challenge.org
http://linkedu.eu/event/lak2013-linkeddata-
tutorial/