Cyberbullying detection and classification with improved IG and BiLSTM
Mengtian Xin, Jiayu Shen, Peifeng Hao · 2022 International Conference on Electronics and Devices, Computational Science (ICEDCS) · 2022
Twitter is a microblogging and social networking service platform; users can post what they feel and think about to share with others. Although it facilitates users' social behaviour, a high degree of freedom of speech also leads to cyberbullying. The statement released by UNICEF showed that 36.5% of middle and high school students experienced cyberbullying, and 87% observed cyberbullying. Cyberbullying has greatly affected people's daily lives. We conduct this study to detect whether online comments contain cyberbullying behaviours and classify cyberbullying to alleviate this problem. This paper uses an improved information gain algorithm for feature selection, and the bidirectional LSTM neural network is used for classification. On the premise that the information gain threshold is limited to 0.0004, the precision on the test set can reach 95.15%.