BhamNLP at SemEval-2020 Task 12: An Ensemble of Different Word Embeddings and Emotion Transfer Learning for Arabic Offensive Language Identification in Social Media

Proceedings of the 14th International Workshop on Semantic Evaluation, pages 1532–1538 Barcelona, Spain (Online), December 12, 2020. 1532 BhamNLP at SemEval-2020 Task 12: An Ensemble of Different Word Embeddings and Emotion Transfer Learning for Arabic Offensive Language Identification in Social Media Abdullah I. Alharbi School of Computer Science University of Birmingham, UK aia784@cs.bham.ac.uk Mark Lee School of Computer Science University of Birmingham, UK m.g.lee@cs.bham.ac.uk Abstract Social media platforms such as Twitter offer people an opportunity to publish short posts in which they can share their opinions and perspectives. While these applications can be valuable, they can also be exploited to promote negative opinions, insults, and hatred against a person, race, or group. These opinions can be spread to millions of people at the click of a mouse. As such, there is a need to develop mechanisms by which offensive language can be automatically detected in social media channels and managed in a timely manner. To help achieve this goal, SemEval 2020 offered a shared task (OffensEval 2020) that involved the detection of offensive text in Arabic. We propose an ensemble approach that combines different levels of word embedding models and transfer learning from other sources of emotion-related tasks. The proposed system ranked 9th out of the 52 entries within the Arabic Offensive language identification subtask. 1 Introduction While microblogging platforms can be used positively and productively, they are also frequently used for destructive purposes such as spreading offensive messages. People who wish to spread hate can use these channels to quickly and easily reach millions of people at the click of a button. To prevent this spread, there is a need for a system that can automatically identify messages that contain offensive language. A significant number of studies have focused on the detection of offensive text in English. However, there is a lack of research that focuses on detecting offensive text in Arabic (Mubarak and Darwish, 2019). To contribute to the emerging body of knowledge in this domain, a shared task ’OffensEval 2020 - Task 12’ was conducted at the SemEval 2020 (Zampieri et al., 2020). One of the subtasks of this shared task is ’Offensive language identification’. The organisers released multilingual datasets for five languages, including Arabic, which is the focus of this paper. The shared task involved some significant challenges. The distribution of the targeted classes was unbalanced: 19% of the tweets were labelled as offensive, while the remaining were labelled as inoffensive. Also, the tweets employed a variety of dialects and sub-dialects. Unlike Modern Standard Arabic (MSA), dialectical Arabic can take a variety of forms, and there is a general lack of rules and standards. To address these challenges, we proposed an ensemble approach that combines three different models (detailed in Section 3). We believe that integrating multiple kinds of classifiers can help overcome the weaknesses and realise the advantages of each. Our proposed system ranked 9th out of 52 participants for the Arabic Offensive language identification subtask. The remainder of this paper is organized as follows. …

Content truncated to 500 words