Using Text Mining to Examine the Role of Topics of Election Promises in Predicting Election Results
The 6th International Conference on Computational Social Science |
Organized by MIT Media Lab
As one of the most popular topics within political communication research, elections
have been paid a lot of attention to, especially the predictors of the election result. Previous
literature has proposed and adequately verified the effect of demographics (e.g. age, gender,
and education), party effect (e.g. liberal and conservative), personal image (e.g. clothing and
smile), political party funding, media strategy, and some other factors of candidates through
the process of election. While, as an important tool for voters to evaluate candidates, form
prospective beliefs about them, and make a voting decision (Born et al., 2018), the election
promise made by candidates through their campaigns and discourses has received less
scholarly attention. Generally, voters have their own policy preferences and candidates who
can implement the preference in their election promises will maximize the chances of
winning the election (Jasim Alsamydai et al., 2013).
Meanwhile, online political information provider like voter guidance website has
improved the openness and transparency in organizing and carrying out elections, which can
improve the engagement of the public (Xenos and Moy, 2007) and make the relationship
between the election promise and result more observable. And the online political
information is the data source of this study.
To identify different policy focuses of candidates, one feasible way is to extract topics
from the election promises of each candidate and quantify them. Compared with traditional
quantitative methods like content analysis, computational methods such as text mining could
save both manpower and time and avoid coder bias to a certain degree. In this study, the
data of candidates of the 2018 local councilor election in Taiwan were used to answer the
following questions:1. How can researchers extract and quantify topics computationally from
election promises? 2. Do policy preferences exist in the local councilor election in Taiwan? If
it exists, which topic of election promises could predict a successful election result?