ACoD: Detecting Abusive Comments from Transliterated Bengali Text Using Deep Learning Techniques

Le résumé fourni par la source

Abusive comment detection refers to the process of identifying and flagging comments or messages in online communication platforms that contain abusive, offensive, or harmful language. Due to inadequate resources and related language processing tools, detecting abusive comments is challenging in low-resource languages like Bengali. The prevalence of transliterated Bengali comments on social media adds to the challenge, as monolingual techniques are unable to capture them. Regrettably, the absence of a publicly accessible, transliterated Bengali corpus for investigating abusive comments highlights a significant gap in available resources. Therefore, this work introduces a new BACoD dataset comprising 6162 transliterated Bengali texts labeled into two categories: abusive and non-abusive. This work explores several machine learning (LR, RF, SVM, MNB) and deep learning (CNN, LSTM, BiLSTM) models for the downstream task and evaluates the developed dataset. Experimental analysis revealed that the CNN+BiLSTM with Keras embedding attained the best performance (W F1score of 0.92).

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.