Towards Safer Social Spaces: LSTM, Bi-LSTM and Hybrid Approach for Cyberbullying Detection in Assamese language on Social Networks

Social media provides a forum for constructive discussion while also providing an environment that is conducive to cyberbullying. Cyberstalking is a growing problem, especially on social media sites like Facebook. It is defined as harassment that occurs online without any physical contact. To mitigate its negative effects, automated procedures for locating and removing cyberbullying content must be put in place. The intent of this research is to broaden the scope of cyberbullying detection, with an emphasis on Assamese. Due to the limited digital resources, Assamese has not received much consideration in previous research. The primary goal of this research is to fill the notable gap in the identification of cyberbullying, specifically in populations who are linguistically marginalised, by focusing on Assamese. Research findings show that LSTM obtained an accuracy of 87.86%, Bi-LSTM attained 89.58% and the hybrid model, LSTM and Bi-LSTM, surpass all with a remarkable accuracy of 89.7%. The research methodology introduces an innovative hybrid approach that melds Long Short-Term Memory (LSTM) models with Bidirectional LSTM (Bi-LSTM), specifically designed to accommodate the unique linguistic characteristics and challenges of Assamese. The results of the research demonstrate the hybrid model's efficacy in identifying cyberbullying in Assamese, indicating its potential utility in improving security when surfing the internet.