Personality-based Deep Learning for Hate Speech Detection

Anonymous

Abstract: An essential factor in the fight against hate speech is the advancement of effective computational algorithms for automatically detecting it. Earlier research has put forth a range of computational methods aimed at automating hate speech detection. However, these approaches have predominantly overlooked significant insights from the psychology literature, which delves into the connection between personality traits and hate. To this end, we propose a novel framework for detecting hate speech focusing on people’s personality factors reflected in their writing. Our framework has two components: (i) a knowledge distillation model for fully automating the process of personality inference from text and (ii) a personality-based deep learning model for hate speech detection. Our approach is unique in that it (i) incorporates low-level personality factors, which have been largely neglected in prior literature, into automated hate speech detection and (ii) proposes multi-head-self-attention-inspired deep learning components for fully exploiting the intricate relationship between personality and hate. In particular, the latter aids the model in untangling intermediate personality factors, the potential existence of which has been suggested by recent research in psychology. We evaluate our model with two real-world datasets. The results show that our model significantly outperforms state-of-the-art baselines. From an academic viewpoint, our study paves the way for future research by incorporating personality aspects into the design of automated hate speech detection. From a business standpoint, our model offers substantial assistance to online social platforms and governmental bodies facing challenges in effectively moderating hate speech.

Paper Type: long

Research Area: Machine Learning for NLP

Contribution Types: NLP engineering experiment, Data analysis

Languages Studied: English

Consent To Share Submission Details: On behalf of all authors, we agree to the terms above to share our submission details.