Abusive Language Detection with Graph Convolutional Networks

Proceedings of NAACL-HLT 2019, pages 2145–2150 Minneapolis, Minnesota, June 2 - June 7, 2019. c⃝2019 Association for Computational Linguistics 2145 Abusive Language Detection with Graph Convolutional Networks Pushkar Mishra⋆, Marco Del Tredici♣, Helen Yannakoudakis♠, Ekaterina Shutova♣ ⋆ Facebook AI, London, United Kingdom ♣ ILLC, University of Amsterdam, The Netherlands ♠ The ALTA Institute, Dept. of CS & Technology, University of Cambridge, United Kingdom pushkarmishra@fb.com, hy260@cam.ac.uk, {m.deltredici, e.shutova}@uva.nl Abstract Abuse on the Internet represents a significant societal problem of our time. Previous re- search on automated abusive language detec- tion in Twitter has shown that community- based profiling of users is a promising tech- nique for this task. However, existing ap- proaches only capture shallow properties of online communities by modeling follower– following relationships. In contrast, working with graph convolutional networks (GCNs), we present the first approach that captures not only the structure of online communities but also the linguistic behavior of the users within them. We show that such a heteroge- neous graph-structured modeling of communi- ties significantly advances the current state of the art in abusive language detection. 1 Introduction Matthew Zook (2012) carried out an interesting study showing that the racist tweets posted in response to President Obama’s re-election were not distributed uniformly across the United States but instead formed clusters. This phenomenon is known as homophily: i.e., people, both in real life and online, tend to cluster with those who ap- pear similar to themselves. To model homophily, recent research in abusive language detection on Twitter (Mishra et al., 2018a) incorporates em- beddings for authors (i.e., users who have com- posed tweets) that encode the structure of their sur- rounding communities. The embeddings (called author profiles) are generated by applying a node embedding framework to an undirected unlabeled community graph where nodes denote the au- thors and edges the follower–following relation- ships amongst them on Twitter. However, these profiles do not capture the linguistic behavior of the authors and their communities and do not con- vey whether their tweets tend to be abusive or not. In contrast, we represent the community of au- thors as a heterogeneous graph consisting of two types of nodes, authors and their tweets, rather than a homogeneous community graph of authors only. The primary advantage of such heteroge- neous representations is that they enable us to model both community structure as well as the lin- guistic behavior of authors in these communities. To generate richer author profiles, we then pro- pose a semi-supervised learning approach based on graph convolutional networks (GCNs) applied to the heterogeneous graph representation. To the best of our knowledge, our work is the first to use GCNs to model online communities in social me- dia. We demonstrate that our methods provide sig- nificant improvements over existing techniques. 2 Related work Supervised learning for abusive language detec- tion was first explored by Spertus (1997) who extracted rule-based features to train their classi- fier. Subsequently, manually-engineered lexical– syntactic features formed the crux of most ap- proaches to the task (Yin …

Content truncated to 500 words