Finding Hidden Swing Voters in the 2022 Italian Elections Twitter Discourse

TL;DR

This work investigates the online political landscape of the 2022 Italian general elections on Twitter by modeling the discourse as time-resolved retweet networks $G_t$ and applying network backboning, community detection, and $k$-core analysis to identify stable, homogeneous political communities. It extends the analysis with a language-based propaganda detector to quantify 18 propaganda techniques and defines voter vulnerability as retweet endorsement of probe tweets, enabling a comparison between swing and non-swing voters. The results show a predominantly stable online discourse with clearly defined political cores, yet reveal measurable swing-voter activity and higher propaganda vulnerability among swing voters, with patterns that vary by swing type and period. The study highlights how popularity, centrality, and core-periphery structure of politicians evolve across campaign phases and demonstrates the nuanced impact of social media on Italian political opinion, offering data- and method-centered insights for researchers and practitioners. The authors provide code to reproduce the analyses, facilitating further exploration of online electoral dynamics in Italy and beyond.

Abstract

The global proliferation of social media platforms has transformed political communication, making the study of online interactions between politicians and voters crucial for understanding contemporary political discourse. In this work, we examine the dynamics of political messaging and voter behavior on Twitter during the 2022 Italian general elections. Specifically, we focus on voters who changed their political preferences over time (swing voters), identifying significant patterns of migration and susceptibility to propaganda messages. Our analysis reveals that during election periods, the popularity of politicians increases, and there is a notable variation in the use of persuasive language techniques, including doubt, loaded language, appeals to values, and slogans. Swing voters are more vulnerable to these propaganda techniques compared to non-swing voters, with differences in vulnerability patterns across various types of political shifts. These findings highlight the nuanced impact of social media on political opinion in Italy.

Figures (9)

• Figure 1: Workflow of the analyses followed in this work.
• Figure 2: Comparison of the Complementary Cumulative Distribution Function (CCDF) for the in-degree distributions across the six retweet networks, split by observation period.
• Figure 3: Party composition of the retweeting communities (with size $\geq20$). Party labels are assigned proportionally to the political representatives in each community. Communities with no political representatives are omitted.
• Figure 4: User migration across communities between consecutive timeframes, with numbers in parentheses indicating the size of each community. Political communities are labeled with the parties included within them. The missing percentage of users primarily results from their inactivity in the second and third periods, as the considered communities include most users.
• Figure 5: K-Core Decomposition of the pre-campaign (left), electoral campaign (center), and post-election (right) networks. Larger and red-colored, resp. smaller and blue-colored, nodes indicate politicians, resp. non-politicians.