Tell Me Who is Your Friend: the Automated Content Analysis of Voters’ Responses to Parties on Social Media During the Electoral Race

The analysis of voters’ perceptions of parties is definitely not a new question in political science and political communication (see Klingemann and Wattenberg 1992, Fernandez-Vazquez 2014, Gorbaniuk et al. 2015, Stubager and Seeberg 2019). However, the traditional methodology usually includes a survey or a variation of a questionnaire (e.g., Trilling 1975, Iyengar and Kinder 1987, Borges and Clarke 2008). Analyzing such data can include an “emotional” lag, i.e. voters are given questions after the elections. Social media presents an opportunity to analyze how voters respond to parties’ statements in the “real time”: political parties often allow their users to leave comments under the Facebook posts. The methodological problem that arises when one wishes to analyze these responses is that it is a daunting task to apply human coding to such data. In this case, the automated text analysis might be a viable solution, with a few promising papers already done in this field (e.g., Vitak et al. 2011, Gustaffson 2012, Ben-David and Matamoros-Fernández 2016, Kalsnes 2016). In this paper, I implement R statistical software to conduct the content analysis of the parties’ Facebook posts, as well as the comments created by users. I am interested to see whether the feedback, given by voters, corresponds to the content presented by the parties. I draw on the Facebook data on four countries - Germany, Switzerland, Austria, and Belguim - during the two most recent consecutive elections. With the use of R packages tm and quanteda, I construct a dataset and implement a sentiment analysis of the voters’ commentaries.