Troll or Abuser Identification and Removal in Social Media for Code-Mixed English, Bangla and Banglish with Machine Learning

Abstract

Social networking platforms have become our center for daily communication. These networks removed all the barriers that used to exist by providing open platform for sharing thoughts, keeping in touch with friends, and even voicing for what we believe in. Unfortunately, this has also opened the door for trolls and online bullies to run rampant. Along with democratizing communication, these platforms have created a distinctive vocabulary, such as the widespread usage of emojis, which are particularly well-liked by younger audiences. ”Banglish” a hybrid language that combines Bengali and English is widely used by Bengalis around the world. It creates serious difficulties for traditional content filtering systems. The study of how people’s views, sentiments, assessments, attitudes, and emotions are conveyed in writing is known as sentiment analysis or opinion mining. This is one of the most active study areas in natural language processing (NPL). Because of its relevance for business and society, this research has expanded beyond computer science and into the management and social sciences. This study gives an overview of the difficulties with sentiment analysis that apply to its methods and procedures. The most recent research that uses deep learning to address issues with sentiment analysis is reviewed in this publication. A comparison study of the experimental findings for various models and input attributes has been carried out. This study examines how to identify trolls and abusive content in Banglish text using NLP models and sophisticated large language models (LLMs), such as OpenAI GPT-4, Google Gemini, and Meta LLaMA. Our goal is to use these models to improve the security and reduce the toxicity of social media environments, ensuring a more secure online experience for all users.