This project found out that Matteo Salvini (the leader of the right-wing party ‘Lega Nord’) uses (consciously or not) emotions to trigger reactions in its voters and, moreover, it shows how, in the years (2011-2018) Matteo Salvini’s staff became aware of how to use social media using formats and refining its copy-writing techniques to stay at the center of the agenda. Even though it is not sure how successful will this strategy be in the elections. This project was pitched by Luca Zorloni and submitted to me, for evaluation, by Andrea Gentile.The process consisted in downloading more than 4k Facebook posts from Matteo Salvini’s page studied with the Syuzhet R package. Such package is used for sentiment analyisis and for the ‘formalistic’ study of novels and other texts.
What makes this project innovative?
This project is innovative because it puts the sentiment analysis of political discourse into a strategic context. What Wired.it asked me to do was not just a description of what was inside Salvini's post, but also to understand what was its purpose. That meant that I had to find some statistical inference about the way sentiment and format (video, text or photos) affect the way a political leader takes the stage on social media. Moreover, even though it was not possible to show it onpage, it required a degree of quantitative statistical analysis to understand the interaction among the variables. In addition, a lot of the data had to be 'created' since Facebook does not provide any form of sentiment analysis.
What was the impact of your project? How did you measure it?
The article had >280 shares on Facebook. Moreover, it was shared by Luca Morisi, Matteo Salvini's spin-doc who appreciated the analysis and by a member of the laboratory of big data and political discourse analysis of the University of Pisa.
Source and methodology
The source was Facebook posts downloaded from the API using my personal token. Once I had downloaded the data from Facebook, I took the post and tokenized them in sentences. The algorithm, then, provided a sentiment analysis of the posts. Once I obtained this result, I started to linearly regress (GLM, LSE and Kernel local linear estimator were used) to understand what caused which posts to receive more likes. I focused on likes because since these data are from before 2015, reactions were not available. Once the quantitative tableau was done, the main task was to find long term trends and to find a theory that would explain everything. This meant to look to academic work on emotions in political discourse.
I coded everything (including the charts) in R. I used the Syuzhet R package for the sentiment analysis and ggplot2 (with ggthemes) for the visualization. In order to download the data from Facebook I had to use the facebookR package.
Luca Zorloni - the guy who had the ideaAndrea Gentile - the guy who edited and supported me during the process