Internet Learning Volume 3, Number 2, Fall 2014 | Page 106
Visualizing Knowledge Networks in Online Courses
Figure 25. Question and Answer Pairs in Renlit Thread.
no other participants mention Media concepts,
and none of the thread’s central concepts
are Media-related. Nonetheless this
thread was judged as onTargetThread=true,
perhaps due to the deep dive into Renlit’s
professional experience, which was also part
of the assignment.
D. RQ3 Discussion
The early topical focus visualization
shown in Figure 23 was time-intensive
and involved a number of manual
steps, but it can serve as a roadmap for automated
approaches. The underlying graph
structure lends itself to automated data extraction
and visualization methods, and can
be used as input to statistical, algorithmic,
machine learning, mathematical, and other
modes of analysis. To illustrate this point,
we will outline a simple, example metric
for calculating individual concept overlap
scores in a discussion.
The conceptOverlap metric emerged
from our desire to somehow quantify the
ways in which participants are connecting
with each other against the backdrop of the
discussion’s concept graph. It is important
to note that in this example we calculate
concept mentions by post, not by author. If
Renlit is the only participant to mention a
concept, but mentions the same concept in
multiple posts, the concept score will still
increment. Properly weighting and interpreting
such factors is an important area for
future work.
The basic formula produces the ratio
of the number of central (multi-connected)
concepts mentioned by a person, to the total
number of concepts mentioned in the conversation.
To state it in graph terms: for a
given author, count the number of concepts
the author mentions where the concept In-
Degree > 1, then divide by the total number
of concepts regardless of InDegree. We
can also produce these ratios with respect to
each concept category, to see how individuals
are contributing to the relative prominence
of central concepts. ConceptOverlap
values for the Renlit thread are provided in
Figure 25.
Upon further testing a score like
conceptOverlap can be adjusted, weighted,
and modified. For example, overlap values
could be weighted depending on the number
of participants mentioning each central
concept, the associated level of topicSpread
or knowledgeActivity, or concept relevance
as determined by comparison with an ontology
of course content.
Also note that conceptOverlap need
not only be measured between individual
posts. For example, it could also be measured
between two individuals over multiple
conversations, between an individual and
the resources they cite, among members of
105