Ferryman at SemEval-2020 Task 12: BERT-Based Model with Advanced Improvement Methods for Multilingual Offensive Language Identification

Proceedings of the 14th International Workshop on Semantic Evaluation, pages 1947–1952 Barcelona, Spain (Online), December 12, 2020. 1947 Ferryman at SemEval-2020 Task 12: BERT-Based Model with Advanced Improvement Methods for Multilingual Offensive Language Identification Weilong Chen, Peng Wang, Jipeng Li, Yuanshuai Zheng, Yan Wang, and Yanru Zhang∗ University of Electronic Science and Technology of China yanruzhang@uestc.edu.cn Abstract Indiscriminately posting offensive remarks on social media may promote the occurrence of negative events such as violence, crime, and hatred. This paper examines different approaches and models for solving offensive tweet classification, which is a part of the OffensEval 2020 competition(Zampieri et al., 2020; Zampieri et al., 2019b). The dataset is Offensive Language Identification Dataset (OLID)(Zampieri et al., 2019a), which draws 14,200 annotated English Tweet comments(Rosenthal et al., 2020). The main challenge of data preprocessing is the unbalanced class distribution, abbreviation, and emoji. To overcome these issues, methods such as hashtag segmentation, abbreviation replacement, and emoji replacement have been adopted for data preprocessing approaches. The main task can be divided into three sub-tasks, and are solved by Term Frequency–Inverse Document Frequency(TF-IDF) vectorizer, Bidirectional Encoder Representation from Transformer (BERT), and Multi-dropout respectively. Meanwhile, we applied different learning rates for different languages and tasks based on BERT and non-BERTmodels in order to obtain better results. Our team Ferryman ranked the 18th, 8th, and 21st with F1-score of 0.91152 on the English Sub-task A, Sub-task B, and Sub-task C, respectively. Furthermore, our team also ranked in the top 20 on the Sub-task A of other languages(C¸ ¨oltekin, 2020; Sigurbergsson and Derczynski, 2020; Mubarak et al., 2020; Pitenis et al., 2020). 1 Introduction With the continuous development of society, online social network (OSN) and microblog sites have attracted Internet users more than any other types of websites. Twitter, Facebook, and Instagram offer a growing variety of services that attract users from different cultures, religions, and interests around the world. The number of OSN users are getting larger, the variety of users and the contents uploaded are growing rapidly. The strong inclusiveness and free atmosphere of social networks provide a platform for all kinds of users to communicate, share, and discuss. Due to the argument caused by different backgrounds, cultures, and beliefs of users, malicious comments have been emerging all these years. The unrestrained release of offensive remarks on social media have negative impacts on the development of the whole society and may lead to outbreaks such as violence, crime, and hatred. The free and inclusive nature of OSN is easy for the widespread of negative effects among large number of users, since the comments and news are almost uncontrollable. Therefore, the detection and protection of offensive content on social media is particularly important. Offensive comments are mostly related to religion, race, and gender, offending, insulting, or threatening others through a series of derogatory words. This is the main part of the objectionable content, which is widely distributed in social media. Many users have abused the power given by social networks to express their opinions excessively. Due to …

Content truncated to 500 words