Cyberbullying Detection Using Bidirectional Encoder Representations from Transformers (BERT)
Razan Sujud, Walid Fahs, Rida Khatoun, Fadlallah Chbib · 2024
Bullying is described as an undesirable behavior by others that harms an individual physically, mentally, or socially. Cyberbullying is a virtual form (e.g., textual or image) of bullying. Thus, detecting cyberbullying behaviors early can prevent long-term psychological harm and promote safer digital spaces. Researchers have increasingly turned to machine learning techniques for effective cyberbullying detection in response to this pressing concern. However, the detection methods that rely on machine learning struggle with contextual understanding. Recently, there has been a shift toward using deep learning models, which have produced novel outcomes. Bidirectional Encoder Representations from Transformers (BERT) specifically utilizes a deep learning approach to learn contextualized representations of words or tokens in a given text corpus. It has been widely used for various Natural Language Processing (NLP) tasks, including text classification. In this paper, we propose a novel approach to building a robust system leveraging BERT, designed to effectively detect and categorize instances of cyberbullying across various online platforms. By employing sophisticated NLP techniques, the objective is to develop a model that can analyze and understand complex contextual details and identify cyberbullying behavior effectively. The system is trained on a diverse collected dataset from different platforms such as YouTube, LinkedIn, and Twitter. Our experimental results demonstrated the ability of our detection model to discriminate between potentially hazardous information and benign interactions according to different performance metrics such as Recall, Precision, and F1-score.