Author(s): Manish Joshi, Dhirendra Pandey, Vandana Pandey and Mohd Waris Khan

Abstract: Cyberbullying in Hinglish, a linguistic fusion of Hindi and English widely utilised on social media, poses considerable issues due to its distinct linguistic features. a number of current detection systems are unable to adequately represent the complexity of Hinglish text, which produces less than ideal outcomes. This paper presents the Hinglish Fusion Framework for Cyberbullying Detection, a method addressing this problem by using advanced natural language processing techniques. The method lets the system detect both semantic and syntactic peculiarities of Hinglish text by combining the contextual strength of a fine-tuned BERT model with the efficiency of FastText embeddings. The framework uses a dual-stream design whereby FastText concentrates on subword-level linguistic information and BERT processes contextual embeddings. Classification performance is improved by a weighted ensemble of outputs derived from these models. Evaluated on a Hinglish cyberbullying dataset, the framework showed notable gains in precision, recall, and F1-score when compared to traditional models. With a scalable and strong solution, more inclusive and efficient moderation tools in multilingual and code-mixed environments are made possible. This paper emphasises the need of hybrid strategies for addressing the difficulties of cyberbullying detection in linguistically varied fields with limited resources.

DOI: 10.33545/26633582.2025.v7.i1a.149

Pages: 07-14 | Views: 1308 | Downloads: 766

Download Full Article: Click Here