USF at SemEval-2019 Task 6: Offensive Language Detection Using LSTM With Word Embeddings

Proceedings of the 13th International Workshop on Semantic Evaluation (SemEval-2019), pages 796–800 Minneapolis, Minnesota, USA, June 6–7, 2019. ©2019 Association for Computational Linguistics 796 USF at SemEval-2019 Task 6: Offensive Language Detection Using LSTM With Word Embeddings Bharti Goel and Ravi Sharma Department of Computer Science and Engineering, University of South Florida, USA Emails: bharti@mail.usf.edu, ravis@mail.usf.edu Abstract In this paper, we present a system description for the SemEval-2019 Task 6 submitted by our team. For the task, our system takes tweet as an input and determine if the tweet is of- fensive or non-offensive (Sub-task A). In case a tweet is offensive, our system identifies if a tweet is targeted (insult or threat) or non- targeted like swearing (Sub-task B). In tar- geted tweets, our system identifies the target as an individual or group (Sub-task C). We used data pre-processing techniques like splitting hashtags into words, removing special charac- ters, stop-word removal, stemming, lemmati- zation, capitalization, and offensive word dic- tionary. Later, we used keras tokenizer and word embeddings for feature extraction. For classification, we used the LSTM (Long short- term memory) model of keras framework. Our accuracy scores for Sub-task A, B and C are 0.8128, 0.8167 and 0.3662 respectively. Our results indicate that fine-grained classification to identify offense target was difficult for the system. Lastly, in the future scope section, we will discuss the ways to improve system per- formance. 1 Introduction In recent years, there has been a rapid rise in social media platforms and surge in the number of users registering in order to communicate, publish con- tent, showcase their skills and express their views. Social media platforms like Facebook and Twit- ter have millions of registered users influenced by the countless user-generated posts on daily basis (Zeitel-Bank and Tat, 2014). While on one hand social media platforms facilitate the exchange of views, effective communication and can be seen as a helping mode in crisis. On the other hand, they open up the window for anti-social behavior such as bullying, stalking, harassing, trolling and hate speech (Malmasi and Zampieri, 2018; Wiegand et al., 2018; ElSherief et al., 2018; Zhang et al., 2018). These platforms provide the anonymity and hence aid users to indulge in aggressive be- havior which propagates due to the increased will- ingness of people sharing their opinions (Fortuna and Nunes, 2018). This aggression can lead to foul language which is seen as “offensive”, “abusive”, or “hate speech”, terms, which are used interchangeably (Waseem et al., 2017). In general, offensive language is de- fined as derogatory, hurtful/ obscene remarks or comments made by an individual (or group) to an individual (or group) (Wiegand et al., 2018; Baziotis et al., 2018). The offensive language can be targeted towards a race, religion, color, gen- der, sexual orientation, nationality, or any char- acteristics of a person or a group. Hate Speech is slowly plaguing the social media users with de- pression and anxiety (Davidson et al., 2017; Zhang et al., 2018), which can be presented in the …

Content truncated to 500 words