Multi-Label Classification using BERT for Cyberbullying Detection
Fidya Farasalsabila, Ema Utami, Suwanto Raharjo · 2024
Cyberbullying is not a new phenomenon in digital issues that even existed before the era of social networks. Cyberbullying has a broad impact, covering a person’s mental and physiological conditions such as sadness, anxiety, and even depression. In the current era of social networks, cyberbullying is increasingly becoming a concern for many groups. Research related to cyberbullying detection is increasingly being carried out. Multi-label datasets are a challenge that must be faced in detecting cases of cyberbullying. Therefore, choosing the right method for the emotion identification process is important. This research analyzes the use of the BERT method on multi-label cyberbullying data. The research uses PyTorch as a machine learning framework to support the research process. The total raw dataset is 47,692 tweets consisting of 6 labels. This algorithm was chosen as a learning solution in the case of sentiment analysis. The results of this study show that BERT can be applied to multi-label classification and achieves the highest accuracy of 94%.