UM-IU@LING at SemEval-2019 Task 6: Identifying Offensive Tweets Using BERT and SVMs

• DOI: 10.18653/v1/s19-2138
• Publication Date: 2019-07-21T17:29:51Z
• Authors: Jian Zhu; Zuoyu Tian; Sandra Kübler
• Keywords: SemEval; Offensive; Macro; F1 score

This paper describes the UM-IU@LING’s system for SemEval 2019 Task 6: Offens-Eval. We take a mixed approach to identify and categorize hate speech in social media. In subtask A, we fine-tuned BERT based classifier detect abusive content tweets, achieving macro F1 score of 0.8136 on test data, thus reaching 3rd rank out 103 submissions. subtasks B C, used linear SVM with selected character n-gram features. For our could target abuse 0.5243, ranking it 27th 65