Author Profiling for Abuse Detection

Proceedings of the 27th International Conference on Computational Linguistics, pages 1088–1098 Santa Fe, New Mexico, USA, August 20-26, 2018. 1088 Author Profiling for Abuse Detection Pushkar Mishra Dept. of CS and Technology University of Cambridge United Kingdom pm576@cl.cam.ac.uk Marco Del Tredici ILLC University of Amsterdam The Netherlands m.deltredici@uva.nl Helen Yannakoudakis Dept. of CS and Technology The ALTA Institute University of Cambridge United Kingdom hy260@cl.cam.ac.uk Ekaterina Shutova ILLC University of Amsterdam The Netherlands e.shutova@uva.nl Abstract The rapid growth of social media in recent years has fed into some highly undesirable phenomena such as proliferation of hateful and offensive language on the Internet. Previous research suggests that such abusive content tends to come from users who share a set of common stereotypes and form communities around them. The current state-of-the-art approaches to abuse detection are oblivious to user and community information and rely entirely on textual (i.e., lexical and semantic) cues. In this paper, we propose a novel approach to this problem that incorporates community-based profiling features of Twitter users. Experimenting with a dataset of 16k tweets, we show that our methods significantly outperform the current state of the art in abuse detection. Further, we conduct a qualitative analysis of model characteristics. We release our code, pre- trained models and all the resources used in the public domain. 1 Introduction Abuse, a term used to collectively refer to offensive language, hate speech, sexist remarks, etc., is om- nipresent in social media. Users on social media platforms are at risk of being exposed to content that may not only be degrading but also harmful to their mental health in the long term. Pew Research Center highlighted the gravity of the situation via a recently released report (Duggan, 2014). As per the report, 40% of adult Internet users have personally experienced harassment online, and 60% have witnessed the use of offensive names and expletives. Expectedly, the majority (66%) of those who have personally faced harassment have had their most recent incident occur on a social networking website or app. While most of these websites and apps provide ways of flagging offensive and hateful content, only 8.8% of the victims have actually considered using such provisions. These statistics suggest that passive or manual techniques for curbing propagation of abusive content (such as flagging) are neither effective nor easily scalable (Pavlopoulos et al., 2017). Consequently, the efforts to automate the detection and moderation of such content have been gaining popularity in natural language processing (NLP) (Waseem and Hovy, 2016; Wulczyn et al., 2017). Several approaches to abuse detection demonstrate the effectiveness of character-level bag-of-words features in a supervised classification setting (Djuric et al., 2015; Nobata et al., 2016; Davidson et al., 2017). More recent approaches, and currently the best performing ones, utilize recurrent neural networks (RNNs) to transform content into dense low-dimensional semantic representations that are then used for classification (Pavlopoulos et al., 2017; Badjatiya et al., 2017). All of these approaches rely solely on lexical and semantic features of the text they are applied to. …

Content truncated to 500 words