Detection of Cyberbullying Incidents on the X Social Network

Social media has transformed communication but also enabled cyberbullying, posing significant psychological risks. The X social network, known for its high volume of real-time interactions and often anonymous user base, is particularly susceptible to cyberbullying incidents. This study uses Natural Language Processing (NLP) and Bidirectional Encoder Representations from Transformers (BERT) to detect cyberbullying on the X social network. We analyzed a Kaggle dataset of 46,017 labeled tweets through data collection, preprocessing, modeling, and evaluation. Preprocessing involved noise removal, tokenization, padding, and numerical conversion. The BERT model, fine-tuned for binary text classification, captured contextual nuances to identify cyberbullying patterns. Evaluation metrics—accuracy, precision, recall, and F1-score—consistently exceeded 85%. Despite hardware limitations, the fine-tuned model showed slight improvements over preset hyperparameters. Future research should leverage more robust resources to enhance accuracy and reliability. This study contributes to minimizing the occurrence of cyberbullying by promoting proactive detection strategies, ultimately fostering safer online interactions