WOLI at SemEval-2020 Task 12: Arabic Offensive Language Identification on Different Twitter Datasets

Proceedings of the 14th International Workshop on Semantic Evaluation, pages 2237–2243 Barcelona, Spain (Online), December 12, 2020. 2237 WOLI at SemEval-2020 Task 12: Arabic Offensive Language Identification on Different Twitter Datasets Yasser Otiefy WideBot y.otiefy@widebot.ai Ahmed Abdelmalek WideBot a.sherif@widebot.ai Islam El Hosary WideBot islam@widebot.ai Abstract Communicating through social platforms has become one of the principal means of personal communications and interactions. Unfortunately, healthy communication is often interfered by offensive language that can have damaging effects on the users. A key to fight offensive language on social media is the existence of an automatic offensive language detection system. This paper presents the results and the main findings of SemEval-2020, Task 12 OffensEval Sub-task A Zampieri et al. (2020), on Identifying and categorising Offensive Language in Social Media. The task was based on the Arabic OffensEval dataset Mubarak et al. (2020). In this paper, we describe the system submitted by WideBot AI Lab for the shared task which ranked 10th out of 52 participants with Macro-F1 86.9% on the golden dataset under CodaLab username "yasserotiefy". We experimented with various models and the best model is a linear SVM in which we use a combination of both character and word n-grams. We also introduced a neural network approach that enhanced the predictive ability of our system that includes CNN, highway network, Bi-LSTM, and attention layers. 1 Introduction Social media platforms such as Facebook and Twitter provide an online space for individuals or groups to express their opinions often with limited restrictions. Hence, social media users may use filthy, sexual, or offensive language Jay and Janschewitz (2008). Offensive phrases could ridicule or insult an individual or a group Razavi et al. (2010). Many applications involve offensive language detection, for instance, parental control or adult content filters; identifying hate speech that incites or presage violence Waseem and Hovy (2016); Identifying troll accounts Darwish et al. (2017); conflict detection, which is often lead up to verbal hostility Chadefaux (2014); quantifying the intensity of polarisation Conover et al. (2011). In addition to the huge effort required by human annotators, the risk of applying discrimination or favouritism should be placed under consideration. Moreover, a manual annotation task by human annotators would increase the system response times drastically, since a computer-based solution can perform this task much faster than humans. With the rapid growth in the amount of user-generated content in the social media services, the scalability of manual filtering highlights the need for automating the process of on-line hate-speech detection Pitsilis et al. (2018). OffensEval, SemEval-2020 Task 12 sub-task A Zampieri et al. (2020) aims to train a model on Arabic tweets collected from Twitter, annotated by human annotators to identify the offensive language within the messages. In this paper, we describe the system submitted by WideBot AI Lab for the shared task of offensive language identification in Arabic and discuss the results of our experiments and it is organised as follows: Section 2 discusses important details about the task and data used. It additionally demonstrates the previous …

Content truncated to 500 words