Analysing Twitter Semantic Networks: the case of 2018 Italian Elections

TL;DR

This paper investigates the semantic networks within Twitter discussions during the 2018 Italian Elections, introducing an automated, behavior-based method to analyze political discourse without manual labeling.

Contribution

It presents a novel, fully automated approach to analyze socio-semantic structures in Twitter debates, grounded in user behavior and applicable across languages and topics.

Findings

Identified discursive communities based on retweeting behavior.

Monitored the evolution of semantic networks over time.

Revealed semantic peculiarities of the Italian electoral debate.

Abstract

Social media play a key role in shaping citizens' political opinion. According to the Eurobarometer, the percentage of EU citizens employing online social networks on a daily basis has increased from 18% in 2010 to 48% in 2019. The entwinement between social media and the unfolding of political dynamics has motivated the interest of researchers for the analysis of users online behavior - with particular emphasis on group polarization during debates and echo-chambers formation. In this context, attention has been predominantly directed towards the study of online relations between users while semantic aspects have remained under-explored. In the present paper, we aim at filling this gap by adopting a two-steps approach. First, we identify the discursive communities animating the political debate in the run up of the 2018 Italian Elections as groups of users with a significantly-similar retweeting behavior. Second, we study the semantic mechanisms that shape their internal discussions by monitoring, on a daily basis, the structural evolution of the semantic networks they induce. Above and beyond specifying the semantic peculiarities of the Italian electoral competition, our approach innovates studies of online political discussions in two main ways. On the one hand, it grounds semantic analysis within users' behaviors by implementing a method, rooted in statistical theory, that guarantees that our inference of socio-semantic structures is not biased by any unsupported assumption about missing information; on the other, it is completely automated as it does not rest upon any manual labelling (either based on the users' features or on their sharing patterns). These elements make our method applicable to any Twitter discussion regardless of the language or the topic addressed.