Generalizable Multilingual Hate Speech Detection on Low Resource Indian Languages using Fair Selection in Federated Learning (ACL Anthology)

Given the linguistic diversity and cultural subtleties of Indian languages, identifying hate speech on social media in these languages is a difficult and hard undertaking. Conventional techniques entail sending sensitive user data to a server for model training, which is undesirable and may provide a privacy risk that has not been well investigated. The study proposes MultiFED, a federated method that effectively detects hate speech, by combining many low-resource language datasets. MultiFED overcomes the constraints of data scarcity by employing ongoing adaptation and fine-tuning to support generalization utilizing subsets of multilingual data. Five distinct pre-trained models are used in extensive tests on 13 Indic datasets. MultiFED surpasses the state-of-the-art baselines by around 8% in terms of accuracy and 12% in terms of F-Score, according to the data.

https://aclanthology.org/2024.naacl-long.400