BNU-HKBU UIC NLP Team 2 at SemEval-2019 Task 6: Detecting Offensive Language Using BERT model

• DOI: 10.18653/v1/s19-2099
• Publication Date: 2019-07-21T17:29:51Z
• Authors: Zhenghao Wu; Hao Zheng; Jianming Wang; Weifeng Su; Jefferson Fong
• Keywords: Offensive; SemEval; Language identification; Identification

In this study we deal with the problem of identifying and categorizing offensive language in social media. Our group, BNU-HKBU UIC NLP Team2, use supervised classification along multiple version data generated by different ways pre-processing data. We then state-of-the-art model Bidirectional Encoder Representations from Transformers, or BERT (Devlin et al, 2018), to capture linguistic, syntactic semantic features. Long range dependencies between each part a sentence can be captured BERT’s bidirectional encoder representations. results show 85.12% accuracy 80.57% F1 scores Subtask A (offensive identification), 87.92% 50% B (categorization offense types), 69.95% 50.47% score C (offense target identification). Analysis shows that distinguishing targeted untargeted is not simple task. More work needs done on unbalance Subtasks C. Some future also discussed.