An Expert Annotated Dataset for the Detection of Online Misogyny

Proceedings of the 16th Conference of the European Chapter of the Association for Computational Linguistics, pages 1336–1350 April 19 - 23, 2021. ©2021 Association for Computational Linguistics 1336 An Expert Annotated Dataset for the Detection of Online Misogyny Ella Guest The Alan Turing Institute University of Manchester ella.guest@manchester.ac.uk Bertie Vidgen The Alan Turing Institute bvidgen@turing.ac.uk Alexandros Mittos Queen Mary University of London University College London alexandros@mittos.net Nishanth Sastry University of Surrey King’s College London The Alan Turing Institute nsastry@turing.ac.uk Gareth Tyson Queen Mary University of London The Alan Turing Institute g.tyson@qmul.ac.uk Helen Margetts The Alan Turing Institute Oxford Internet Institute hmargetts@turing.ac.uk Abstract Online misogyny is a pernicious social prob- lem that risks making online platforms toxic and unwelcoming to women. We present a new hierarchical taxonomy for online miso- gyny, as well as an expert labelled dataset to enable automatic classification of misogynistic content. The dataset consists of 6,567 labels for Reddit posts and comments. As previous research has found untrained crowdsourced an- notators struggle with identifying misogyny, we hired and trained annotators and provided them with robust annotation guidelines. We report baseline classification performance on the binary classification task, achieving accu- racy of 0.93 and F1 of 0.43. The codebook and datasets are made freely available for fu- ture researchers. 1 Introduction Misogyny is a problem in many online spaces, mak- ing them less welcoming, safe, and accessible for women. Women have been shown to be twice as likely as men to experience gender-based online harassment (Duggan, 2017). This misogyny can inflict serious psychological harm on women and produce a ‘silencing effect’, whereby women self- censor or withdraw from online spaces entirely, thus limiting their freedom of expression (Mantilla, 2013; International, 2017). Tackling such content is increasingly a priority for social media platforms and civil society organisations. However, detecting online misogyny remains a difficult task (Hewitt et al., 2016; Nozza et al., 2019). One problem is the lack of high-quality datasets to train machine learning models, which would enable the creation of efficient and scal- able automated detection systems (Anzovino et al., 2018). Previous research has primarily used Twitter data and there is a pressing need for other platforms to be researched Lynn et al. (2019a). Notably, de- spite social scientific studies that show online miso- gyny is pervasive on some Reddit communities, to date a training dataset for misogyny has not been created with Reddit data. In this paper we seek to address the limitations of previous research by presenting a dataset of Reddit content with expert labels for misogyny that can be used to develop more accurate and nuanced classification models. Our contributions are four-fold. First, we de- velop a detailed hierarchical taxonomy based on existing literature on online misogyny. Second, we create and share a detailed codebook used to train annotators to identify different types of miso- gyny. Third, we present a dataset of 6,383 entries from Reddit. Fourth, we create baseline classifi- cation models based on these datasets. All of the research artefacts are made …

Content truncated to 500 words