Misogyny and Aggressiveness Tend to Come Together and Together We Address Them

Proceedings of the 13th Conference on Language Resources and Evaluation (LREC 2022), pages 4142–4148 Marseille, 20-25 June 2022 © European Language Resources Association (ELRA), licensed under CC-BY-NC-4.0 4142 Misogyny and Aggressiveness Tend to Come Together and Together We Address Them Arianna Muti, Francesco Fernicola and Alberto Barr´on-Cede˜no Department of Interpreting and Translation Universit`a di Bologna Forl`ı, Italy arianna.muti2@unibo.it francesco.fernicola@studio.unibo.it a.barron@unibo.it Abstract We target the complementary binary tasks of identifying whether a tweet is misogynous and, if that is the case, whether it is also aggressive. We compare two ways to address these problems: one multi-class model that discriminates between all the classes at once: not misogynous, non aggressive-misogynous and aggressive-misogynous; as well as a cascaded approach where the binary classification is carried out separately (misogynous vs non-misogynous and aggressive vs non-aggressive) and then joined together. For the latter, two training and three testing scenarios are considered. Our models are built on top of AlBERTo and are evaluated on the framework of Evalita’s 2020 shared task on automatic misogyny and aggressiveness identification in Italian tweets. Our cascaded models —including the strong na¨ıve baseline— outperform significantly the top submissions to Evalita, reaching state-of-the-art performance without relying on any external information. Keywords: misogyny identification, aggressiveness identification, social media analysis 1. Introduction Even if Twitter was conceived to express personal opin- ions, share big events, follow people or just communi- cate with friends, an increasing number of users misuse it by engaging in trolling, cyberbullying, or by posting aggressive and misogynous contents (Samghabadi et al., 2020). According to Nocentini et al. (2010), these contents feature an imbalance of power, intention, repe- tition, anonymity and publicity. Anonymity in particular has allowed the amount of hateful posts to dramatically increase. As a consequence, Twitter (as other platforms) struggles to control inappropriate contents.1 Accord- ing to Vox (Italian Rights Observer), women are more targeted than ever on Twitter. Although the overall num- ber of hateful tweets decreased in 2020, the number of misogynous tweets increased significantly: from 26% to 49%.2 Misogynous contents express hating towards women, in the form of insults, sexual harassment, male privilege, patriarchy, gender discrimination, belittling, violence, body shaming and sexual objectification (Sri- vastava et al., 2017). A misogynous text turns aggressive when it overtly or covertly encourages or legitimizes violence against women. Since both Fulper et al. (2014) and Blake et al. (2021) showed that the amount of misogynous tweets is strongly correlated with rape and violence, employing automatic tools to identify them might limit those events. The development of tools to flag hateful language au- tomatically —in particular towards women, since they are subjected to more bullying, abuse, hateful language, and threats than men— is a must (Fallows, 2005). 1https://business.twitter.com/en/ help/ads-policies/ads-content-policies/ inappropriate-content.html 2http://www.voxdiritti.it/ la-nuova-mappa-dellintolleranza-5/ For these reasons, we approach the problem of spot- ting whether a tweet in Italian is misogynous and, if it is, whether it is also aggressive. We specifically fo- cus on detecting aggressiveness in misogynous tweets. Existing systems struggle to identify the target of an aggressive statement, leading …

Content truncated to 500 words