Emad at SemEval-2019 Task 6: Offensive Language Identification using Traditional Machine Learning and Deep Learning approaches

Proceedings of the 13th International Workshop on Semantic Evaluation (SemEval-2019), pages 600–603 Minneapolis, Minnesota, USA, June 6–7, 2019. ©2019 Association for Computational Linguistics 600 Emad at SemEval-2019 Task 6: Offensive Language Identification using Traditional Machine Learning and Deep Learning approaches Emad Kebriaei1, Samaneh Karimi2, Nazanin Sabri3, Azadeh Shakery4 1234School of Electrical and Computer Engineering, College of Engineering, University of Tehran, Iran 4School of Computer Science and Institute for Research in Fundamental Sciences (IPM) {emad.kebriaei,samanekarimi,nazaninsabri,shakery}@ut.ac.ir Abstract In this paper, the used methods and the results obtained by our team, entitled Emad, on the OffensEval 2019 shared task organized at Se- mEval 2019 are presented. The OffensEval shared task includes three sub-tasks namely Offensive language identification, Automatic categorization of offense types and Offense target identification. We participated in sub- task A and tried various methods including traditional machine learning methods, deep learning methods and also a combination of the first two sets of methods. We also pro- posed a data augmentation method using word embedding to improve the performance of our methods. The results show that the augmen- tation approach outperforms other methods in terms of macro-f1. 1 Introduction With the growth of social networking platforms, the need for automatic methods that manages the emerging issues or facilitate using them is rising. One of the rising trends in social networks such as Twitter is offensive behavior that can cause the of- fended users leave their social network. Therefore, the need for effective automatic methods for iden- tifying offensive language in textual data is impor- tant. The OffensEval shared task has been organized in order to give a boost to computational methods for identifying and categorizing offensive content on social media. Three sub-tasks defined in the of- fensEval shared task are identification of offensive language(sub-task A), categorization of offense types(sub-task B) and identification of the offense target(sub-task C) (Zampieri et al., 2019b). The main goal in sub-task A is to identify of- fensive tweets from non-offensive ones. By defi- nition, a post is labeled as offensive if it contains any form of non-acceptable language (profanity) or a targeted offense, which can be veiled or di- rect. This year, we participated in sub-task A. Our methods for this sub-task include two approaches. In the first approach, traditional machine learning methods, deep learning methods and also a com- bination method are employed for the task. In the second approach, a data augmentation method is proposed to improve the performance of the meth- ods of the first approach. 2 Related Work Offenseive language identification which is also known as aggression, cyberbullying, hate speech and abusive language has been widely stud- ied in previous works(Davidson et al., 2017; Malmasi and Zampieri, 2017, 2018). Based on a survey conducted by Fortuna and Nunes (2018), the majority of previous works are on English and the researchers mainly use machine learning for this task and most proposed methods on abusive content detec- tion have modeled the problem as a binary classification task. Based on another survey (Schmidt and Wiegand, 2017), …

Content truncated to 500 words