Automatic hate speech detection: a case study on online comments with a focus on self-victimisation and sentiment

Abstract This study investigates the correlation of hate speech and self-victimisation by focusing on below-the-line (BTL) comments on Ukrainians in Polish news articles and blogs. The comments analysed were posted between 20/04/2020 and 19/04/2021 and therefore before the Russian invasion in February 2022. The analysis was based on two corpora, one representative and one tailored corpus, and a two-tier methodology with a quantitative and a qualitative phase. The quantitative phase was conducted on Sketch Engine, and it focused on identifying and extracting a self-victimisation trope through the lexemes we-they, victim, Pole, Ukrainian . The qualitative phase focused on the BTL comments’ general sentiment towards Ukrainians. Two software packages were used, MultiEmo and HateSpeech (CLARIN-PL), and the respective results were compared. The findings show that on average 30 % of the comments feature negative sentiment, but not hate speech per se, however the self-victimisation trope is notable and worthy of further investigation. Recommendations are provided to fine-tune the software and increase manual annotation in follow-up studies.