IITP-AINLPML at SemEval-2020 Task 12: Offensive Tweet Identification and Target Categorization in a Multitask Environment

Proceedings of the 14th International Workshop on Semantic Evaluation, pages 1983–1991 Barcelona, Spain (Online), December 12, 2020. 1983 IITP-AINLPML at SemEval-2020 Task 12: Offensive Tweet Identification and Target Categorization in a Multitask Environment Soumitra Ghosh, Asif Ekbal, Pushpak Bhattacharyya Department of Computer Science and Engineering Indian Institute of Technology Patna, India {1821cs05, asif, pb}@iitp.ac.in Abstract In this paper, we describe the participation of IITP-AINLPML team in the SemEval-2020 Shared Task 12 on Offensive Language Identification and Target Categorization in English Twitter data. Our proposed model learns to extract textual features using a BiGRU-based deep neural network supported by a Hierarchical Attention architecture to focus on the most relevant areas in the text. We leverage the effectiveness of multitask learning while building our models for sub-task A and B. We do necessary undersampling of the over-represented classes in the sub-tasks A and C. During training, we consider a threshold of 0.5 as the separation margin between the instances belonging to classes OFF and NOT in sub-task A and UNT and TIN in sub-task B. For sub-task C, the class corresponding to the maximum score among the given confidence scores of the classes (IND, GRP and OTH) is considered as the final label for an instance. Our proposed model obtains the macro F1-scores of 90.95%, 55.69% and 63.88% in sub-task A, B and C, respectively. 1 Introduction With the advancement of technology and gaining popularity of several social media platforms, there has been a manifold increase in the number of users active in such channels. More often than not, these users tend to misuse their power of freedom of speech and violate acceptable usage policies of most of the forums. This has necessitated detecting any offensive or obscene posts, comments, images, etc. and prevent further dissemination of the same to curtail its effect on social media. One promising solution is to develop automated tools for identification of offensive language using various Natural Language Processing (NLP) and Machine Learning (ML) techniques. In the past few years, many shared tasks and seminars were introduced to address this problem and provide relevant annotated data collected from various social media. Some of the related tasks are: ALW1 (related to Abusive Language Identification), TRAC 1 2018 2 related to Aggression Identification (Kumar et al., 2018), GermEval Task 2 3, HatEval 2019 4(i Orts, 2019), HASOC 2019 5 and OffensEval 2019 Task 6 6 related to Offensive Language Identification (Zampieri et al., 2019b). Recent works have considered categorizing hate speech problem into sub-classes like abusive, aggressive, or offensive speech. Such categorization of social media posts help law-enforcement agencies with the surveillance of social media. 1.1 Problem definition Similar to OffensEval-2019 shared task, the OffensEval-2020 shared task-organized at SemEval-2020 involves detection of Offensive Language and Target Identification from Tweets in English. OffensEval- 2020 (Zampieri et al., 2020) offers a multilingual (Arabic, Danish, English, Greek, Turkish) dataset 1https://sites.google.com/site/abusivelanguageworkshop2017/ 2https://sites.google.com/view/trac1/shared-task 3https://projects.fzai.h-da.de/iggsa/ 4https://competitions.codalab.org/competitions/19935 5https://hasocfire.github.io/hasoc/2019/index.html 6https://sites.google.com/site/offensevalsharedtask/offenseval2019 This work is licensed under a Creative Commons Attribution 4.0 International Licence. Licence details: http:// creativecommons.org/licenses/by/4.0/. 1984 …

Content truncated to 500 words