Hate Speech Detection Based on Sentiment Knowledge Sharing

Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on Natural Language Processing, pages 7158–7166 August 1–6, 2021. ©2021 Association for Computational Linguistics 7158 Hate Speech Detection based on Sentiment Knowledge Sharing Xianbing Zhou1, Yong Yang1, Xiaochao Fan1∗, Ge Ren1, Yunfeng Song1, Yufeng Diao2, Liang Yang2, Hongfei Lin2∗ 1School of Computer Science and Technology, Xinjiang Normal University, China 2Department of Computer Science and Technology, Dalian University of Technology, China {1783696285, 68523593, 37769630, 236789497, 1697277502}@qq.com {diaoyufeng, liang}@mail.dlut.edu.cn, hflin@dlut.edu.cn Abstract The wanton spread of hate speech on the in- ternet brings great harm to society and fami- lies. It is urgent to establish and improve au- tomatic detection and active avoidance mecha- nisms for hate speech. While there exist meth- ods for hate speech detection, they stereotype words and hence suffer from inherently biased training. In other words, getting more affec- tive features from other affective resources will significantly affect the performance of hate speech detection. In this paper, we propose a hate speech detection framework based on sentiment knowledge sharing. While extract- ing the affective features of the target sentence itself, we make better use of the sentiment fea- tures from external resources, and finally fuse features from different feature extraction units to detect hate speech. Experimental results on two public datasets demonstrate the effective- ness of our model. 1 Introducon With the prevalence of mobile Internet and social media, phenomena such as the malicious spread of hate speech have gradually become widespread. This often has incalculable consequences and has become a serious social problem. How to quickly and accurately detect hate speech automatically, and then better intervene to prevent it has become one of the hot research issues in the field of nat- ural language processing. The automatic detec- tion of hate speech can prevent the viral spread of hate speech, thereby reducing the malicious spread of cyberbullying and harmful information. In the field of public opinion analysis, monitoring and intervention, hate speech detection has extensive value in application. In recent years, the hate speech detection has been paid more attention, and many research re- sults have appeared. However, the task is quite ∗corresponding author: Xiaochao Fan, Hongfei Lin. challenging due to the inherent complexity of the natural language constructs. Most of the existing works revolves either around rules (Krause and Grassegger, 2016) or manual feature extraction (Gitari et al., 2015). Rule-based methods do not in- volve learning and typically rely on a pre-compiled list or dictionary of subjectivity clues (Haralam- bous and Lenca, 2014). Chen et al. (2012) pro- posed a variety of linguistic rules to determine whether a sentence constitutes hate speech or not. For example, if a second-person pronoun and a derogatory word appear at the same time, such as “<you, gay>”, the sentence is judged to be insult- ing. This type of method not only requires manual formulation of rules, but also requires dictionaries of derogatory words. There have also been many attempts to detect hate speech …

Content truncated to 500 words