Mitigating Bias in Session-based Cyberbullying Detection: A Non-Compromising Approach

Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on Natural Language Processing, pages 2158–2168 August 1–6, 2021. ©2021 Association for Computational Linguistics 2158 Mitigating Bias in Session-based Cyberbullying Detection: A Non-Compromising Approach Lu Cheng1∗, Ahmadreza Mosallanezhad1∗, Yasin N. Silva2, Deborah L. Hall3, and Huan Liu1 1 Computer Science and Engineering, Arizona State University 2 Mathematical and Natural Sciences, Arizona State University 3 Social and Behavioral Sciences, Arizona State University {lcheng35, amosalla, ysilva, d.hall, huanliu}@asu.edu Abstract The element of repetition in cyberbullying be- havior has directed recent computational stud- ies toward detecting cyberbullying based on a social media session. In contrast to a single text, a session may consist of an initial post and an associated sequence of comments. Yet, emerging efforts to enhance the performance of session-based cyberbullying detection have largely overlooked unintended social biases in existing cyberbullying datasets. For exam- ple, a session containing certain demographic- identity terms (e.g., “gay” or “black”) is more likely to be classified as an instance of cyber- bullying. In this paper, we first show evidence of such bias in models trained on sessions col- lected from different social media platforms (e.g., Instagram). We then propose a context- aware and model-agnostic debiasing strategy that leverages a reinforcement learning tech- nique, without requiring any extra resources or annotations apart from a pre-defined set of sensitive triggers commonly used for identify- ing cyberbullying instances. Empirical evalu- ations show that the proposed strategy can si- multaneously alleviate the impacts of the un- intended biases and improve the detection per- formance. 1 Introduction Cyberbullying has become a prevalent adverse be- havior in online social interactions. Recent find- ings indicate that over 35% of young people have been victims of cyberbullying and roughly 15% have admitted to cyberbullying others (Hinduja and Patchin, 2020; Kim et al., 2021). The detrimental consequences of cyberbullying have motivated con- siderable efforts in various fields to combat cyber- bullying. For example, in computational studies of cyberbullying detection – which have been largely aimed at classifying text posted on social media ∗Equal contribution 0.2 0.4 0.6 0.8 P(label = bully|z) Density of the Conditional Probability z = without sensitive z = sensitive (a) HAN. 0.0 0.2 0.4 0.6 0.8 1.0 P(label = bully|z) Density of the Conditional Probability z = without sensitive z = sensitive (b) Debiased HAN. Figure 1: Conditional probability densities of standard HAN and debiased HAN on sessions with and without sensitive triggers z in the Instagram dataset released by (Hosseinmardi et al., 2015). platforms with machine learning and natural lan- guage processing (NLP) – the primary goal is to improve the overall accuracy and speediness of de- tection. Partly due to an increased awareness of the repetitive nature of cyberbullying behavior, a number of recent efforts in cyberbullying detection have shifted in focus from classification of a sin- gle text to detection in a social media session. A session typically consists of an image/video with a caption, a sequence of comments, …

Content truncated to 500 words