A Comparative Study of Cyberbullying Detection Using Fine-Tuned Large Language Models
Shriyan Shekhar · 2024
Cyber Bullying is a problem that plagues the online environment. With the advent of social media after Covid-19, there has been a surge in the number of cases of cyberbullying. Cyberbullying has long-term impacts on the minds of students. There needs to be accurate cyberbullying identification methods. However, there are multiple obstacles when it comes to identifying cyberbullying as the language structure is complex and there are changing terminologies. It is time to leverage a fast-paced system that can understand the complexities of language; the best solution to this is large language models. Sophisticated models like Mistral 7B, Vicuna 7B, and LLaMA 7B can be important in identifying cyberbullying. Training these models using Low-Rank Adaptation (LoRA) and Reinforcement Learning from Human Feedback (RLHF) can lead to cyberbullying detection with high accuracy. Through extensive testing, this research shows the capabilities of LLMs to detect cyberbullying. This research aims to protect vulnerable groups, foster a safe environment, and automate the recognition of cyberbullying with greater precision.