Moral Values in Social Media for Disinformation and Hate Speech Analysis

Social networks face criticism for their links to disinformation and hate speech but offer unprecedented research opportunities to contrast them. This work focuses on three categories in the Italian social dialogue: political entities (parties or politicians), reliable news outlets and questionable news outlets. Social media behavioural differences emerge between these categories when including moral information in analysing tweet production and their responses. We created a dataset of over 175,000 tweets on immigration covering a 5-year period and enriched it with reliability annotation and toxicity scores. Also, we exploited a neural network model to label tweets according to the Moral Foundations Theory. We found significant relations between moral information, unreliability, engagement, and toxicity score, allowing us to interpret those behaviours. These relations were analysed over time for tweets sorted by moral content, and significant differences in the distribution of production and toxicity levels emerged between the categories. This result and the analysis of similarities between the accounts based on moral expressions and community engagement showed that the accounts categories have distinct behaviours, demonstrating the importance of moral information in assessing the news and political debate in social media.