1.2 Computational text analysis complements existing methods
It is a staggering challenge to collect, interpret, and prioritize the
vast literature on Arctic system change - and then synthesize this
information into scenarios about the future. Systematic meta-analysis of
Arctic futures work (Arbo et al., 2013) has revealed the utility of
synthetic review, but such meta-analysis may also be too narrowly
focused on relatively small sample sizes (e.g., less than 100 articles).
Others have argued that there is simply too much content to consume,
and, as a result, pre-existing knowledge and biases are employed to
filter content (Schatzmann et al., 2013). Thus, there is a need for
complimentary approaches that can scan the information across hundreds
or thousands of documents and across a broad range of disciplines to
accurately understand the trajectory of where the world is heading (Kwon
et al., 2017).
In thinking about the future Arctic, it is necessary to think well
outside-the-box regarding nonlinear changes in technology and global
change (Dator, 1993, 2019; Johnson, 2011). One method for addressing
biased selection of evidence can be to develop a large corpus of
literature that captures a very broad range of content related to the
future Arctic. However, human cognitive biases actively disregard
information that seems outlandish or strange (Schoemaker, 2004), making
unbiased interpretation a persistent challenge.
Computational text analysis can assist in addressing this bias by
providing a complimentary means of thematic identification and analysis
of a corpus of literature. In other words, a computer program can be
used to potentially reduce human selection bias by ‘reading’ the
thousands of texts and ‘interpreting’ different patterns than a human
might see. Specifically, machine learning-based latent Dirichlet
allocation can enable the rapid analysis of large, text-based datasets
to reveal thematically distinct clusters of information (Asmussen &
Møller, 2019).
This approach has been used in a limited way for scenarios related to
drone technology (Kwon et al., 2017) and electric cars (Kim et al.,
2016). Considerable opportunity exists to leverage these methods and
fully incorporate them into detailed visions of the future Arctic
(Kayser & Blind, 2017; Kayser & Shala, 2016).