Hate Speech and Counter Speech Detection: Conversational Context Does Matter

Proceedings of the 2022 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, pages 5918 - 5930 July 10-15, 2022 ©2022 Association for Computational Linguistics Hate Speech and Counter Speech Detection: Conversational Context Does Matter Xinchen Yu University of North Texas xinchenyu@my.unt.edu Eduardo Blanco Arizona State University eduardo.blanco@asu.edu Lingzi Hong University of North Texas lingzi.hong@unt.edu Abstract Hate speech is plaguing the cyberspace along with user-generated content. This paper in- vestigates the role of conversational context in the annotation and detection of online hate and counter speech, where context is defined as the preceding comment in a conversation thread. We created a context-aware dataset for a 3-way classification task on Reddit com- ments: hate speech, counter speech, or neutral. Our analyses indicate that context is critical to identify hate and counter speech: human judg- ments change for most comments depending on whether we show annotators the context. A linguistic analysis draws insights into the lan- guage people use to express hate and counter speech. Experimental results show that neural networks obtain significantly better results if context is taken into account. We also present qualitative error analyses shedding light into (a) when and why context is beneficial and (b) the remaining errors made by our best model when context is taken into account. 1 Introduction The advent of social media has democratized public discourse on an unparalleled scale. Meanwhile, it is considered a particularly conducive arena for hate speech (Caiani et al., 2021). Online hate speech is prevalent and can lead to serious consequences. At the individual level, the victims targeted by hate speech are frightened by online threats that may materialize in the real world (Olteanu et al., 2018). At the societal level, it has been reported that there is an upsurge in offline hate crimes targeting mi- norities (Olteanu et al., 2018; Farrell et al., 2019). There are two common strategies to combat on- line hate: disruption and counter speech. Disrup- tion refers to blocking hateful content or users. To scale this strategy, researchers have proposed methods to identify hate (Waseem and Hovy, 2016; Davidson et al., 2017; Nobata et al., 2016). While these interventions could de-escalate the impact of Parent As an average height male, idgaf how tall you are, if that’s your issue then spend the money and get a better seat, or just f**king make the seat selection online to get more space. Target Found the short guy! -Target is Neutral if considering only Target. -Target is Hate if considering Parent and Target. Parent I deal with women all day with my job and this is how they are - extremely stupid, hate-filled, bizarre and they appreciate nothing. Target Maybe you’re an a**hole if they treat you like that? -Target is Hate if considering only Target. -Target is Counter-hate if considering Parent and Target. Table 1: Reddit comments (Targets) deemed to be Hate, Neutral, or Counter-hate depending on whether one takes into account the previous comment (Parent). hate speech, they may …

Content truncated to 500 words