“Are you kidding me?”: Detecting Unpalatable Questions on Reddit

Proceedings of the 16th Conference of the European Chapter of the Association for Computational Linguistics, pages 2083–2099 April 19 - 23, 2021. ©2021 Association for Computational Linguistics 2083 “Are you kidding me?”: Detecting Unpalatable Questions on Reddit Sunyam Bagga .txtLAB McGill University sunyam.bagga@mcgill.ca Andrew Piper Dept. of Languages, Literatures, & Cultures McGill University andrew.piper@mcgill.ca Derek Ruths School of Computer Science McGill University derek.ruths@mcgill.ca Abstract Abusive language in online discourse nega- tively affects a large number of social me- dia users. Many computational methods have been proposed to address this issue of online abuse. The existing work, however, tends to focus on detecting the more explicit forms of abuse leaving the subtler forms of abuse largely untouched. Our work addresses this gap by making three core contributions. First, inspired by the theory of impoliteness, we pro- pose a novel task of detecting a subtler form of abuse, namely unpalatable questions. Sec- ond, we publish a context-aware dataset for the task using data from a diverse set of Reddit communities. Third, we implement a wide ar- ray of learning models and also investigate the benefits of incorporating conversational con- text into computational models. Our results show that modeling subtle abuse is feasible but difficult due to the language involved be- ing highly nuanced and context-sensitive. We hope that future research in the field will ad- dress such subtle forms of abuse since their harm currently passes unnoticed through exist- ing detection systems. 1 Introduction Abusive language and other antisocial behaviour is omnipresent in online discourse. According to a recent survey, 41% of Americans have person- ally experienced some form of online harassment (Duggan, 2017). To counter abusive behaviour on- line, different social media platforms implement their own mechanisms such as content moderation, muting or blocking users from posting etc. It is, however, infeasible to manually moderate online communities due to the sheer enormity of content produced every day – Twitter, for example, receives over 500 million tweets per day. Manual moder- ation in such a scenario would require humans to read millions of tweets daily which would take an impractical amount of time and other resources. Consequently, many computational models have been proposed by the Natural Language Process- ing (NLP) community to detect online abuse and facilitate automatic content moderation. Abuse is an umbrella term which can cover several types of negative expressions. There ex- ists a plethora of abuse detection studies employ- ing different terminology: personal attacks (Wul- czyn et al., 2017), bullying (Dadvar et al., 2013; Chatzakou et al., 2017), hate speech (Warner and Hirschberg, 2012; Davidson et al., 2017; Djuric et al., 2015; Gao and Huang, 2017), nastiness (Samghabadi et al., 2017), harassment (Golbeck et al., 2017; Yin et al., 2009), hostility (Liu et al., 2018), racism or sexism (Waseem and Hovy, 2016), abusive language (Nobata et al., 2016), aggression (Caines et al., 2018), and others. However, extant work in abuse detection has largely focused on de- tecting overt abuse ignoring the more subtle forms of abuse which can …

Content truncated to 500 words