Internet Learning Volume 3, Number 2, Fall 2014 | Page 78

Visualizing Knowledge Networks in Online Courses
• Question: Does a discussant ask a
question?
• Personal Story: Does a discussant tell
a story from personal experience?
• Citation: Does a discussant make
reference to a book, article, or other
work (citation)?
• Challenge: Does a discussant challenge
another discussant?
D. Task Target and On-Targetness
To understand conversations in our
formal learning environment, we also
felt it important to consider the targeted
behavior of the collaborative activity
or discussion prompt. Activities (discussion
prompts) were coded using the knowledge-
Activity and topicSpread categories. For example,
tasks might ask students to Transfer
and Elaborate (knowledgeActivity=2/Transfer,
topicSpread=3/Elaborate). General topical
alignment was also considered.
Each discussant’s comment, as well
as the entire thread, was coded for whether
or not it was on target in relation to the original
task prompt. These binary attributes are
called onTargetPost, and onTargetThread.
E. Metadata Attributes
Finally, we identified a set of quantitative
attributes that provide more information
about individual participants
as well as the shape and structure of conversations
themselves. These included:
• word count (of participants, conversations,
and individual responses)
• number of posts (for each participant
and conversation)
• number of unique participants (in
each conversation)
• time stamp (of each participant’s posts
and the conversation as a whole)
• proximity of posts in time (of each
participant’s posts and for the conversation
as a whole)
• level of the response tree at which a
response is posted (responseLevel)
F. Intersectionality
We believed that our richest insights
from this type of exploratory
study would spring from
our ability to identify and visualize the intersection
of individual, conversational and
content characteristics. For example, do
certain combinations of individual students
generate more ‘productive’ or ‘successful’
conversations? Are student and instructor
questions treated differently? What kinds
of instructor strategies might be effective in
various kinds of conversations? How does
the introduction of certain concepts or resources
impact the depth or number of participants
in a conversation? See Figure 2 for
some examples of these intersectionalities.
With this emergent framework as
our guide, we manually coded a data set of
948 threaded discussion posts for targeted
attributes; designed a graph schema and
graph database to aid in describing and analyzing
the problem space; and began the
project of designing queries and visualizations
to facilitate analysis of the threaded
discussion data from graph computing and
Natural-Language Processing (NLP) perspectives.
G. Tools Development and Scalability
We decided to employ or build
technology solutions where feasible,
but to not limit our questions
to what was possible with current
technologies. We favored a data design that
would speak well to our questions, even if at
first it would require significant labor to op-
77