Improved Hierarchical Attention Networks for Cyberbullying Detection via Social Media Data

Wenhao Qian, Songnan Yu, Zheng Nie, Xiaoyu Sean Lu, Haoyue Liu, Bo Huang

Abstract: Nowadays, online bullying, called cyberbullying, has become one of the most common problems faced by young people in society. The text, containing and spreading amounts of information, is a key medium in the social media and social networks. The natural language processing (NLP) method as the most common method deals with textual information. In the field of cyberbullying detection, extracting information from texts is a way to predict online bullying phenomenon presently. In this work, we propose a transformer based hierarchical neural network (THNN). Compared to existing methods, ours has the following advantages: (i) it captures longer context information for longer time intervals; (ii) it reflects and mimics the hierarchical structure of conversation media clearly; (iii) it has less computational time.