Enhancing Hate Speech Detection through Emoji-based Classification using Bi-LSTM and GloVe Embeddings

The emergence of user-friendly and affordable online networking platforms like YouTube and Instagram have provided people of all ages with a platform to express themselves and share their personal experiences. Regrettably, this has also resulted in surge of harmful and unwarranted statements being made on these platforms, with the aim of damaging someone’s reputation or social status. To combat this, it is crucial for both the government and these platforms to acknowledge and take steps to curb hate speech. However, identifying hate speech automatically can be a daunting task, particularly in tech-savvy countries like India, where non-standard spellings, varied grammatical structures, and code-mixed language with emojis are prevalent. In this study, we delve into the identification of hateful speech within text that incorporates emojis through the implementation of machine learning and deep learning models on code-mixed social media content. Our approach involves analyzing approximately 42,000 tweets that feature both textual and emoji-based features. The results of our recommended approach have been favorable in detecting abusive language in relevant datasets.