“In this study, we construct large-scale tweet datasets for hate speech detection in English and a low-resource language, Turkish, consisting of human-labeled 100k tweets per each. Our datasets are designed to have equal number of tweets distributed over five domains. The experimental results supported by statistical tests show that Transformer-based language models outperform conventional bag-of-words and neural models by at least 5% in English and 10% in Turkish for large-scale hate speech detection. The performance is also scalable to different training sizes, such that 98% of performance in English, and 97% in Turkish, are recovered when 20% of training instances are used. We further examine the generalization ability of cross-domain transfer among hate domains. We show that 96% of the performance of a target domain in average is recovered by other domains for English, and 92% for Turkish. Gender and religion are more successful to generalize to other domains, while sports fail most.” https://arxiv.org/pdf/2203.01111.pdf Share this: Click to print (Opens in new window) Print Click to share on Facebook (Opens in new window) Facebook Click to share on LinkedIn (Opens in new window) LinkedIn Click to share on Reddit (Opens in new window) Reddit Click to share on WhatsApp (Opens in new window) WhatsApp Click to share on Bluesky (Opens in new window) Bluesky Click to email a link to a friend (Opens in new window) Email Like this:Like Loading... Post navigation Impact and dynamics of hate and counter speech online (Springer Open) L3Cube-MahaHate: A Tweet-based Marathi Hate Speech Detection Dataset and BERT models (arXiv)