"Stylistic analysis of a hate speech corpus"
by Dr Antonio Pascucci, L’Orientale University of Naples
Abstract
The hate speech phenomenon is a cybercrime that has been growing in recent years. Hate speech, although a sociological phenomenon, has its full realization through written and spoken texts. On the internet, and in social networks in particular, people are more likely to adopt aggressive behaviour because of the anonymity provided by these environments (Brunap and Williams, 2015). Social media represent a sort of echo chamber, in which more radical expressions than those of face-to-face interaction are used. The NLP community is at the forefront in developing AI systems for hate speech detection on the web. Despite this, Fortuna and Nunes (2018) emphasise the need to use multi-class approaches (based on different hate speech categories) instead of only binary classification (e.g. hate speech vs. non-hate speech, misogyny vs. non-misogyny and so on). For this reason, I carried out research on a hate speech corpus in order to investigate stylistic differences in different categories of hate speech (e.g. racism, hate based on religion, LGBTQI+phobia, misogyny). My aim was to show that it is possible to distinguish between i) hate speech and non-hate speech texts and ii) hate speech categories by focusing on haters’ writing style.
Speaker’s bio
Antonio Pascucci is a research assistant at the L’Orientale University of Naples and a member of the UNIOR NLP Research Group. He received his PhD from the same university in 2022. He is well known for his work in computational stylometry, hate-speech detection and authorship attribution. he is also a member of the COST Action 17124 “Digital Forensics. Evidence Analysis via Intelligent Systems and Practices”. He has published in conferences and workshops such as ACII, CLiC-it and TRAC. Furthermore, he serves in the programming committee of many AI and NLP workshops targeting author profiling in abusive language such as ResT-UP.