Safeguarding the Integrity of Online Social Networks (OSN): Leveraging the Efficacy of Conv-LSTM-Based Siamese Network to Predict Hate Speech in Low Resource Hindi-English Code-Mixed Text

Abstract

In the era of free speech and rapid internet expansion, curbing the dissemination of offensive content on social media has become a pressing concern for linguists and regulatory bodies. Hate speech not only targets individuals or groups but also impacts their mental well-being, often leading to feelings of anxiety, isolation, and helplessness. Therefore, hate speech detection should be viewed not just as a linguistic task but also as a public health imperative. In multilingual and culturally diverse countries like India, the challenge is heightened by the presence of code-mixed language. While most existing studies focus on monolingual data, our work addresses hate speech detection in Hindi-English code-mixed text. We propose a Convolution-LSTM network that incorporates spatial and temporal features. Furthermore, the model’s performance is enhanced by constructing an ensemble network using an early-fusion-based Siamese architecture. Experimental results demonstrate that our approach outperforms existing baselines in identifying hate speech in low-resource, code-mixed scenarios.