Geography Reference
In-Depth Information
computational methods such as latent Dirichlet allocation (LDA) (Wallach
2006
)
can be used to comb thousands of posts to reveal key themes in course discussions.
Finally, we use a combined named-entity recognition (NER) and geocoding
approach (Karimzadeh et al.
2013
) to extract and visualize the places that students
mention in their discussion posts. The results of this work provide important
insights for cartography educators, as they reveal key motivations for students in
a globally-diverse MOOC to use, make, and understand maps. These insights may
then feed into future iterations of cartography courses of all sizes and delivery
modes.
Background
At a broad level, our research is concerned with the general problem associated with
making sense out of large text collections. Visualizing text is possible through a
wide range of means, including methods such as tag clouds (Hassan-Montero and
Herrero-Solana
2006
), which can provide a general overview of the frequency of
terms, and more sophisticated methods that use self-organized maps to develop
spatialized representations to show the topics that appear in a text collection, and to
expose their relative similarity across text datasets (Skupin and Fabrikant
2003
).
The former approach leverages a visual technique alone, with minimal computa-
tional effort required. The latter approach requires sophisticated computational
methods to identify and extract patterns, before visualization becomes possible.
Our work leverages data developed by students taking the first MOOC to focus
on cartographic design (hereafter referred to as the Maps MOOC). The Maps
MOOC featured 5 weeks of lessons on the most general cartographic competencies.
Students learned how to recognize spatial thinking, understand basic spatial ana-
lyses, and apply core cartographic design principles. The activities of the more than
48,000 students who enrolled to take the class were logged in a variety of ways, and
the datasets that result constitute large, complex datasets in the context of distance
education. Characterizing what can be uncovered from large textual datasets is a
common contemporary problem for information visualization (D¨rk et al.
2010
)
and geographic visualization researchers (MacEachren et al.
2011
), and the Maps
MOOC offers a very large text dataset in the form of over 95,000 discussion forum
posts. An advantage of this dataset for cartographic inquiry is that it
s reasonable to
expect a large proportion of this discussion to be grounded in discussions about
Geography, and therefore lend itself to geovisualization research.
To begin making sense of this large and diverse textual data, we apply the use of
three complementary methods in this paper. In the following sections we apply one
relatively simple visual method in the form of Phrase Nets to evaluate statements
students contributed regarding how they currently use maps. We then explore how
the computational approach of topic modeling through latent Dirichlet allocation
can be used to mine discussion forum posts to reveal major topics that students
discussed. Finally, we make use of a modified named-entity extraction method to
'
