Internet Learning Volume 3, Number 2, Fall 2014 | Page 88
Visualizing Knowledge Networks in Online Courses
Visually identifying difference
might also allow instructors to more easily
target messaging and feedback to individual
students. Figure 21 illustrates the relatively
regimented participation patterns of
the instructors in our data set, as compared
with the more free-flowing timing of student
contributions. Instructor corpora are
also strikingly similar to each other, as compared
with the diversity of student corpora.
Though we cannot be sure of the reason for
this regimented behavior, it is safe to suggest
that as class sizes increase, it becomes
difficult simply to read the massive volume
of student contributions, much less to fairly
assess contributions or craft individualized
responses. Corpus diagrams could help instructors
in large online courses by presenting
high-level summaries and signifiers to
help them target attention, participate more
effectively, and perhaps gauge the effectiveness
of various response interventions over
time and at scale. These visualizations can
not only provide instructors with a better
understanding of student contributions,
but also perhaps provide students and instructors
with tools for perceiving, assessing,
and focusing their own behaviors and
interaction strategies.
Although there is not enough space
to discuss it here, we have also experimented
with creating a ‘concept corpus’
for each participant. This model connects a
person directly to the concepts mentioned
throughout their response corpus, producing
a concept graph of that person’s favored
discussion topics over time, which could be
used to recommend content, connect with
peer tutors, or form effective work groups.
8.3. RQ3 Example describes the construction
of a concept graph for a discussion
thread, and Interactive 4 allows basic exploration
of that concept graph. This example
can be used to imagine how an individual
concept corpus could be utilized.
VII - RQ2 Findings: Can we identify,
differentiate, and visualize conversation attributes
and behaviors in an online discussion
or course?
A. RQ2 Conceptual Overview
In 6. RQ1 FINDINGS, we considered a
collection of hand-coded response attributes
across a discussant corpus as a
means of representing, differentiating, and
reasoning about individual discussants,
using digital-ethnographic readings as an
analytical anchor. The individual corpus, as
the unit of analysis, was constructed based
on the relations between a person and their
associated response nodes in the graph.
What makes that analysis possible is consistent
and replicable corpus generation based
on the underlying structure of the graph.
Individual corpora may vary, but their underlying
structural properties are the same.
Now, in 7. RQ2 FINDINGS, we
investigate the interactional, influential,
temporal, and co-creational aspects of individuals
participating in discussion threads.
We approach this problem using the same
SKN attributes and digital-ethnographic
descriptions, mapped onto the somewhat
more complex graph structures of threaded
discussion response trees. We also describe
a graph-structural influence metric, called
DiscussionRank, that can be used to gauge
the impact of a response, response author,
particular speech act, or other event on the
evolution of a discussion thread.
B. RQ2 Technical Summary
For conversation modeling, we can use
the Discussion--contains-->Response
and Response--hasResponse-->Response
relations in our schema to extract
a basic subgraph of the desired discussion.
87