Projects

CONTROVERSY PREDICTION

This project aims at identifying controversial news topics which may elicit toxic reactions by different on-line communities. Controversies are best defined as lengthy conversations between different actors that whose remain unchanged over the course of time and become more and more polarized. Controversies are not a bad thing per se, but they may easily shift into toxic debates where offenses and insults can easily occur among the involved actors, thus restricting participation to debates in the public sphere rather than promoting it. Being able to automatically identify potential controversial topics, or events, can help moderators and social media managers to prevent the generation of toxic environments.

Currently, we are conducted research on this topic in Italian and in Dutch, using data from Facebook and Reddit.

HATE SPEECH

The aim of this project is to detect typical patters of online hate speech by exploiting machine learning models.

Hate speech is a controversial and broad topic, hence, the study's objective is to employ longitudinal data extracted from platforms of different nature: social media, newspapers, and blogs. Different sources caters to diverse users, whose interpretation of events is the interest of this study.

The first iteration of the research is slated to cover:

1. Longitudinal data collection: merging existing corpora and newly manually annotated data - NOW -
2. Model Development
3. Study of the results and search for Hate Speech patterns
4. Model Improvement