SINAI at SemEval-2020 Task 12: Offensive Language Identification Exploring Transfer Learning Models

Proceedings of the 14th International Workshop on Semantic Evaluation, pages 1622–1627 Barcelona, Spain (Online), December 12, 2020. 1622 SINAI at SemEval-2020 Task 12: Offensive language identification exploring transfer learning models Flor Miriam Plaza-del-Arco, M. Dolores Molina-Gonz´alez, L. Alfonso Ure˜na-L´opez, M. Teresa Mart´ın-Valdivia Department of Computer Science, Advanced Studies Center in ICT (CEATIC) Universidad de Ja´en, Campus Las Lagunillas, 23071, Ja´en, Spain {fmplaza, mdmolina, laurena, maite}@ujaen.es Abstract This paper describes the participation of SINAI team at Task 12: OffensEval 2: Multilingual Offensive Language Identification in Social Media. In particular, the participation in Sub-task A in English which consists of identifying tweets as offensive or not offensive. We preprocess the dataset according to the language characteristics used on social media. Then, we select a small set from the training set provided by the organizers and fine-tune different Transformer- based models in order to test their effectiveness. Our team ranks 20th out of 85 participants in Subtask-A using the XLNet model. 1 Introduction In recent years the way of sharing information has suffered a change by the worldwide proliferation of use of social media. However, the freedom of this use often leads many users to communicate using offensive language. Offensive language is defined as the text which uses hurtful, derogatory or obscene terms made by one person to another person (Wiegand et al., 2019). One of the strategies used to deal with offensive behavior on social media is to monitor or report this type of comment by users. However, this strategy is not entirely feasible due to the huge amount of data generated daily by users. Therefore, it is necessary to develop automatic systems capable of identifying this type of content on the Web. In this paper, we present the systems we developed as part of our participation in SemEval-2020 Task 12: OffensEval 2 - Multilingual Offensive Language Identification in Social Media (Zampieri et al., 2020). In particular, the participation in Sub-task A in English which consists of identifying tweets as offensive or not offensive. The rest of the paper is structured as follows. In Section 2 some previous related studies are introduced. In Section 3, we describe the pre-trained models we have studied to address the task. In Section 4 we explain the data used in our methods, we present the details of the proposed systems and we discuss the analysis and evaluation results for our system. Finally, we conclude in Section 5 with remarks. 2 Background In recent years, the textual analysis of offensive detection has attracted the attention of several researchers in the field of Natural Language Processing (NLP) leading to various academic events and shared tasks in different languages such as the first, second and third editions of the Workshop on Abusive Language (Roberts et al., 2019); the first and second edition of GermEval Shared Task on the Identification of Offensive Language (Struß et al., 2019); the shared task on aggression identification included in the First Workshop on Trolling, Aggression and Cyberbullying (Kumar et al., 2018); the MEX-A3T track …

Content truncated to 500 words