MIDAS at SemEval-2019 Task 6: Identifying Offensive Posts and Targeted Offense from Twitter

Proceedings of the 13th International Workshop on Semantic Evaluation (SemEval-2019), pages 683–690 Minneapolis, Minnesota, USA, June 6–7, 2019. ©2019 Association for Computational Linguistics 683 MIDAS at SemEval-2019 Task 6: Identifying Offensive Posts and Targeted Offense from Twitter Haimin Zhang,1 Debanjan Mahata,1 Simra Shahid,2 Laiba Mehnaz,2 Sarthak Anand,3 Yaman Kumar,5 Rajiv Ratn Shah,4 Karan Uppal1 1Bloomberg, USA, 2DTU-Delhi, India, 3NSIT-Delhi, India, 4IIIT-Delhi, India, 5Adobe, India hzhang449@bloomberg.net, dmahata@bloomberg.net, simrashahid bt2k16@dtu.ac.in, laibamehnaz@dtu.ac.in, sarthaka.ic@nsit.net.in, ykumar@adobe.com, rajivratn@iiitd.ac.in, kuppal8@bloomberg.net Abstract In this paper, we present our approach and the system description for Sub-task A and Sub Task B of SemEval 2019 Task 6: Identify- ing and Categorizing Offensive Language in Social Media. Sub-task A involves identify- ing if a given tweet is offensive or not, and Sub Task B involves detecting if an offensive tweet is targeted towards someone (group or an individual). Our models for Sub-task A is based on an ensemble of Convolutional Neu- ral Network, Bidirectional LSTM with atten- tion, and Bidirectional LSTM + Bidirectional GRU, whereas for Sub-task B, we rely on a set of heuristics derived from the training data and manual observation. We provide a de- tailed analysis of the results obtained using the trained models. Our team ranked 5th out of 103 participants in Sub-task A, achieving a macro F1 score of 0.807, and ranked 8th out of 75 participants in Sub Task B achieving a macro F1 of 0.695. 1 Introduction The unrestricted use of offensive language in so- cial media is disgraceful for a progressive society as it promotes the spread of abuse, violence, ha- tred, and leads to other activities like trolling. Of- fensive text can be broadly classified as abusive and hate speech on the basis of the context and tar- get of the offense. Hate speech is an act of offend- ing, insulting or threatening a person or a group of similar people on the basis of religion, race, caste, sexual orientation, gender or belongingness to a specific stereotyped community (Schmidt and Wiegand, 2017; Fortuna and Nunes, 2018). Abu- sive speech categorically differs from hate speech because of its casual motive to hurt using gen- eral slurs composed of demeaning words. Both of them are the popular categories of offensive con- tent, widespread in different social media chan- nels. With the democratization of the web, the usage of offensive language in online platforms is a clear indication of misuse of our right to ‘Freedom of Speech’. While censorship of free moving online content curtails the freedom of speech, but unreg- ulated opprobrious tweets discourage free discus- sions in the virtual world making the problem of identifying and filtering out offensive content from social media an important problem to be solved for creating a better society, both in and out of the In- ternet. Detecting offensive content from social media is a hard research problem due to variations in the way people express themselves in a linguis- tically diverse social setting of the web. A major challenge in monitoring online content produced on social media …

Content truncated to 500 words