Internet Learning Volume 3, Number 2, Fall 2014 | Page 84
Visualizing Knowledge Networks in Online Courses
B. RQ1 Technical Summary
A
corpus diagram requires a graph
containing a person, and all associated
responses. We collected those
responses by following all outgoing ‘wrote’
edges from a given person, as follows:
g.V.has(‘personName’,’Renlit’).
out(‘wrote’)
We wrote the results to an in-memory
Tinkergraph, exported the data as
GraphML, and imported to Gephi for further
modeling. We applied a consistent set
of visualization rules, such as node sizing
based on wordCount, and color mappings
for the values of various attributes. Finally,
we applied a force-directed graph layout
algorithm to the model to obtain a readable
presentation. Based on that model, we
used Gephi to export a separate SVG vector
graphics file for each attribute’s color
scheme, and overlaid them using Adobe
Illustrator. As a final step, we exported to
PDF format while preserving top-level Illustrator
layers, resulting in a layered PDF.
We used these PDFs as data analysis tools,
and to generate the comparative corpus diagrams
presented in this paper.
C. RQ1 Example
The following figures use comparative
corpus diagrams, coded for a handful
of attributes, to illustrate a few similarities
and distinctions among three participants:
Renlit and Loret, who are students,
and Naya, who is a lead course instructor.
Each corpus diagram represents the entire
history of each discussant’s contributions
over multiple weeks and courses, and is accompanied
by a brief description of the participant
based on our digital-ethnographic
observation data. A brief comparison will
illustrate how elements of these participants’
digital-ethnographic descriptions can be detected
using comparative corpus diagrams,
and the potential of the approach to support
identification and differentiation of individuals
based on their patterns of discussion
participation.
Figure 5 compares corpora for Renlit,
Loret, and Naya, coded for usage of personalStories.
Renlit’s diagram shows the
highest level of story usage across the entire
data set, and reflects the digital-ethnographic
description of Renlit’s tendency to answer
questions using personalStories rooted in a
professional context. Loret shows story usage
at a significantly lower level than Renlit,
but more in line with typical student
numbers. Naya, on the other hand, uses
only one personalStory in a corpus of 91 responses,
the largest corpus in the data set.
Naya’s responses are significantly shorter
than most student responses, with an average
wordCount of 61. We can’t infer that all
instructors in all situations will show such
a marked difference from students in this
regard, but in combination with other data
points, these provide a promising starting
point for differentiating participants.
Figure 6, coded for questions, reveals
a striking correlation between Naya’s corpus
diagram, and the digital-ethnographic
description of Naya as favoring short, probing
questions as a participation strategy. A
comparison of Naya with Loret and Renlit
is also revealing. For stories, Renlit was prolific
and Naya barely registered, with a gap
of about 70%. For questions, the situation
is flipped, with Naya asking many questions
and Renlit asking relatively few, with a gap
of approximately 50%. And in both cases,
Loret is in between, in some cases appearing
more like the other student, and in some
appearing more instructor-like, as reflected
in the digital-ethnographic description.
83