Natural Language Processing for Cyberbullying Detection

Abstract

With the development of digital technologies and the popularity of social media, cyberbullying has become a serious public health concern that can lead to increased risk of mental and behavioral health issues or even suicide. Artificial intelligence like machine learning opens a lot of possibilities to combat cyberbullying, e.g. automatic cyberbullying detection. Most recent research focuses on improving performance by developing complex models that demand more resources and time to run. The research uses publicly available datasets without carefully evaluating their feasibility and limitations. This study uses natural language processing (NLP) to evaluate the model performance and examine the difference between fine-grained classification and binary classification as well as assess the feasibility and quality of the publicly available dataset. The results show that simple classifier can also achieve similar performance as that of more complex models if appropriate preprocessing is used, and the publicly available dataset may have limitations and quality issues that researchers should consider when using the data.