Project PredicHate

Pragmatic modelling for preventing antisocial behaviour

About the project

Digital platforms facilitate communication and interaction, but they can also contribute to the emergence and escalation of hostile behaviours. The PredicHate project aims to better anticipate such developments by analysing not only the words used, but also their context: who is addressing whom, with what intention, and within what conversational dynamics. In particular, the project will examine implicit meanings, participants’ roles, and strategies that may either fuel or, conversely, help de-escalate tensions. This information will be structured into an ontology and used to annotate a corpus of online conversations. The project will then develop graph-based models capable of jointly representing messages, users, and their relationships. The aim is to design more accurate and explainable systems for identifying conversational trajectories that may lead to antisocial behaviour.