Cyberbullying detection based on aspect-level sentiment analysis
Tong Pan · 2024
Cyberbullying is prevalent among groups such as adolescents and college students, posing a significant threat to mental health. The advancement of natural language processing technology has enabled the rapid and effective detection of cyberbullying language, facilitating real-time monitoring of online harassment on the Internet. Therefore, researching how to better identify cyberbullying language holds important social significance. This study employs aspect-level sentiment analysis methods to achieve fine-grained recognition of cyberbullying language considering textual orientation. It combines named entity recognition to extract aspect words and proposes an aspect word sentiment analysis model based on the BERT-ADA model. The model enhances its learning capability for domain-specific knowledge through in-domain retraining. The results reveal that the proposed approach exhibits optimal model performance compared to popular baseline sentiment analysis models. Additionally, named entity recognition and in-domain retraining significantly enhance the model's performance.