Issues and Engagement in Italian Election Posts on Facebook in 2018 and 2022

And the final (!) session at the P³: Power, Propaganda, Polarisation ICA 2024 postconference starts with the great Giada Marino, presenting today on the work of the Vera.ai research project, which seeks to develop AI tools to monitor and combat mis- and disinformation on social media. This part of that project examines digital traces on Facebook during the 2018 and 2022 Italian elections.

Italy has a multi-party political system with a variety of parties. The 2018 election was won by the populist Five Star Movement, governing together with the far-right Lega. In 2022, a coalition of right-wing and far-right parties including Lega, Fratelli d’Italia, and others won the election. The present study builds on the work of my team that pointed out that not all polarisation is problematic, but that we need to focus instead on instances of destructive polarisation – and of the five symptoms we identified focusses especially on the amplification of extreme voices (and marginalisation of moderate perspectives), and on exclusion as driven by emotional responses that deepens social divides.

Some of this can be operationalised by noting that on Facebook, the seeking of consonant political information is consistently associated with anger towards political opponents and enthusiasm towards like-minded partisans – through emotions expressed in comments, as well as ‘angry’ reactions towards counter-attitudinal content and ‘love’ reactions and content on-sharing for attitude-congruent content. Posts from hyperpartisan accounts are likely to receive more such positive reactions from attitudinally aligned users.

But compiling the datasets required for such work is a challenge, especially when working in languages other than English. Gathering relevant posts on political communication in Italian on Facebook is difficult, and there is a general lack of exposure data for the content of such posts; the platform does not usually make such data available.

Some of this was addressed here through the use of the Facebook URL Shares dataset. This shows 68 million URLs circulated on Facebook between 2017 and October 2022, from 46 countries, with broad user engagement data available for these; this is a user-centric dataset which covers URLs that were shared publicly at least 100 times. The dataset can be filtered for engagement predominantly by Italian users.

Using this dataset, the study examined whether the news sources’ ideological positioning was associated with specific emotional responses. During the 2018 and 2022 election, some 65,000 and 20,000 distinct URLs circulated predominantly amongst Italian users, respectively; of these, a smaller subset related to political topics (as determined by a binary classifier).

The texts of these articles were then clustered by their word embedding, and GPT-4 was used to label these clusters. They were then analysed using the Multi-Party Media Partisanship Attention Score (MP-MPAS) approach, and assessed for their level of emotional polarisation. Immigration-related issues emerged as particularly prominent in both elections; 2018 was dominated by supportive as well as antagonist narratives towards the Five Star Movement; 2022 news was also dominated by energy cost concerns; vaccination, the education system, and political scandals were also prominent in both years.

Five Star Moment campaigning generated more love than anger in 2018; elite financial privilege and scandals generated more anger. In 2022, anger was directed in the far right towards parliamentary pensions and public sector salary caps, on the left, towards issues around worker abuse. The MP-MPAS coverage in 2022 is fairly low, making an interpretation of issue clusters more difficult.