Facebook debate on Sea Watch 3 case: detecting offensive language through Automatic Topic Mining Techniques

Over the years, there has been growing concern about the disproportionate
use of hate speech on social media platforms. In this paper, we
present a text analysis for detecting abusive language in Italian messages on
Facebook, surrounding the debate over the migrant-rescue ship, Sea Watch 3,
and its captain Carola Rackete. The study data consists of more than 130,000
posts retrieved from two pages relating to Matteo Salvini, the leader of the
Italian Lega political party, and from the official Facebook pages of five Italian
newspapers. To explore the presence of offensive and hatred expressions
in the corpus and to establish to what extent social users’ language differs,
depending on the type of Facebook pages analysed, we ran a topic model
based on Latent Dirichlet Allocation. We have complemented this approach
with tools from semantic network analysis