Bilingual hate speech detection on social media: Amharic and Afaan Oromo (Journal of Big Data)

Four distinct deep learning classifiers (CNN, BiLSTM, CNN-BiLSTM, and BiGRU) and three feature extraction methods (Keras word embedding, word2vec, and FastText) were used in this study to detect hate speech in both Amharic and Afaan Oromo. The experiment shows that BiLSTM with FastText feature extraction outperforms the other approach, detecting hate speech in Bilingual Amharic Afaan Oromo with an accuracy of 78.05%. The out-of-vocabulary (OOV) issue is resolved by the FastText feature extraction. Additionally, we are trying to make the resource available to support future study in the field of bilingual hate speech detection for other under-resourced Ethiopian languages and to incorporate other linguistic aspects of the languages to detect hate speech.

https://journalofbigdata.springeropen.com/articles/10.1186/s40537-024-01044-y