Analysing Moral Beliefs for Detecting Hate Speech Spreaders on Twitter

Mirko Lai, Marco Antonio Stranisci, Cristina Bosco, Rossana Damiano, Viviana Patti

Abstract: The Hate and Morality (HaMor) submission for the Profiling Hate Speech Spreaders on Twitter task at PAN 2021 ranked as the 19th position - over 67 participating teams - according to the averaged accuracy value of $$73\%$$ over the two languages - English ( $$62\%$$ ) and Spanish ( $$84\%$$ ). The method proposed four types of features for inferring users attitudes just from the text in their messages: HS detection, users morality, named entities, and communicative behaviour. In this paper, since the test set is now available, we were able to analyse false negative and false positive prediction with the aim of shed more light on the hate speech spreading phenomena. Furthermore, we fine-tuned the features based on users morality and named entities showing that semantic resources could help in facing Hate Speech Spreaders detection on Twitter.