Exploring affiliation network models as a collaborative filtering mechanism in elearning

1
Exploring affiliation network models as a
collaborativefiltering mechanism in e-learning

• Miguel-Angel Sicilia
• Salvador Sánchez-Alonso
• Leonardo Lezcano

Information Engineering Research Unit Computer
Science Dept., University of Alcalá
2
AOP relationships
People work with such services, for learning
about a particular objective (e.g. topic,
competency).
Nowadays learning experiences are setup around
activities.
Learners interact through different kinds of
services like newsgroups and chats.
This kind of relationship activity-objective-peopl
e (AOP) is the basic material for the empirical
analysis of social interaction through technology
enhanced learning.
Analyzing Methods
Quantitative analysis
Qualitative analysis

• Computing of actual social interaction indicators
• Help tutors in decision making
• Processes large amounts of communication events
• Social Network Analysis

• Intensive effort from the tutors to categorize
  and examine each of the interventions
• Exposed to subjectivity of tutors

3
Social Network Analysis (SNA)
General purposes in e-learning

• Hypothesis testing or exploratory studies aimed
  to finding correlations
• The summative assessment of learners

• Re-configuring the learning environment or
  undertaking other actions based on the analysis
  data

Concretely, we approach AOP data in the form of
an affiliation network, considering that
learners participation in activities can be used
to detect groups of common interest. Also,
modeling data in that way, make it possible to
devise different forms of collaborative
filtering
The usual interpretation of collaborative
filtering is that of recommendations or ranking
of information. Here we adopt a more general
position, considering collaborative filtering as
any course of action taken on the basis of the
analysis of the social network structure.
4
Affiliation Network
Two disjoint sets defining a bipartite network
Required Preconditions
Threads planned time must be similar
Discussion threads (events)
Tutors or learners (actors)
Participation should not be made mandatory
Each thread must have a clear topic or objective,
distinguishable from the rest
Undirected ties that affiliate actors with events
The above preconditions guarantee, to the extent
possible, that participation in discussion is a
function of interest, so the more the learner
contributes to discussing a topic, the more
she/he shows an interest in the topic, thus
allowing for a form of quantitative indicator
5
Filtering participants

• The affiliation network can be used to implement
  different strategies for the definition of
  subgroups.
• Identify groups that are close or distant in
  their interests.
• Turn student groups into effective teams
  (Oakley, 2004)

Test structural equivalence, (actors that have
similar relations to the others) with
block modelling technique, which provides a way
of doing this with the help of automated
algorithms.
Compute the participation of actors in each of
the topics, and then examine relationships with a
hypergraph.
6
Blockmodeling technique
Processing Steps

• Remove tutors and nodes with degree lower than
  two
• Randomize learners and topics order
• Set the number of partitions depending on the
  number of learners and topics (lt6,6gt)
• Apply Random Block Modeling

Learners
Main Features

• Able to detect different kind of structures (e.g.
  cohesion, centrality)
• Allows exceptions or errors on input data (e.g.
  Empirical data)
• Effective only for small dense networks

Topics
7
Instructor-led on-the-fly filtering
Very active learners that show low interest in
practical topics (computer tools)
Partitions of learners with no significant
activity

• Combining different interests to foster
  discussion or combining the same interest to
  better focus those discussions.
• Combining more active and more passive groups, or
  filter out the latter.

Active learners that show low interest in
theoretical issues
This group shows attention only to introductory
issues on e-learning
In general there is less interest from topics T5
onwards Introduce reinforcement activities
8
Changing course structure

• Re-organize structure joining or splitting
  topics.
• Topics that are connected with a high strength
  can be joined together, or even be separated in
  another course.
• Concepts that are more peripheral might be
  removed, separated or re-arranged for future
  editions of the same learning experience.

Enhanced Modularity
Therefore we need to identify highly related
topics to a given intensity and well get it with
the help of m-slices.
9
m-Slices
One-Mode valued network
The larger the edge value between two topics the
stronger or more cohesive the common interest
4-slice
16-slice
33-slice
An m-slice is a maximal subnetwork
containing the lines with a multiplicity equal to
or greater than m and the vertices incident with
these lines.
Colours show the nesting of the slices.
Yellow ones are also red and red ones are also
blue.
m-slice are nested
10
T4H2 and T4H4 are about IMS LD and poorly related
to the rest, so it could be reasonable to
separate LD contents to a second part of the
course
4-slice
33-slice is cohesive group of interest that
includes the three Introduction Topics, so they
could be joined together
16-slice
33-slice
T6 is about IEEE LOM and it is closely related to
the rest.
11
Conclusions
Use of affiliation models for exploring on-line
interaction in e-learning
Development of mathematical, quantitative
techniques for filtering the environment
Because there arent clear-cut thresholds for
automated structure settings, tutor should take
described techniques as an indicator to aid in
decision making over the learning process
Further Work

• Evaluate indicators, regarding AOP data and their
  potential usages.
• Gather evidence to turn them into standard
  facilities in e-learning platforms.
• Provide an advanced tool for the analysis of
  social interaction.