KEIS@JUST at SemEval-2020 Task 12: Identifying Multilingual Offensive\n Tweets Using Weighted Ensemble and Fine-Tuned BERT

2020/05/15 by Saja Khaled Tawalbeh, Tawalbeh, Saja Khaled, M. Hammad +3
Computer Science · #Hate Speech and Cyberbullying Detection

paper · pdf · doi:10.48550/arxiv.2005.07820

Abstract

This research presents our team KEIS@JUST participation at SemEval-2020 Task\n12 which represents shared task on multilingual offensive language. We\nparticipated in all the provided languages for all subtasks except sub-task-A\nfor the English language. Two main approaches have been developed the first is\nperformed to tackle both languages Arabic and English, a weighted ensemble\nconsists of Bi-GRU and CNN followed by Gaussian noise and global pooling layer\nmultiplied by weights to improve the overall performance. The second is\nperformed for other languages, a transfer learning from BERT beside the\nrecurrent neural networks such as Bi-LSTM and Bi-GRU followed by a global\naverage pooling layer. Word embedding and contextual embedding have been used\nas features, moreover, data augmentation has been used only for the Arabic\nlanguage.\n

Related

• UPB at SemEval-2020 Task 12: Multilingual Offensive Language Detection on Social Media by Fine-tuning a Variety of BERT-based Models
• KUISAIL at SemEval-2020 Task 12: BERT-CNN for Offensive Speech\n Identification in Social Media
• LT@Helsinki at SemEval-2020 Task 12: Multilingual or language-specific\n BERT?
• LIIR at SemEval-2020 Task 12: A Cross-Lingual Augmentation Approach for\n Multilingual Offensive Language Identification
• ANDES at SemEval-2020 Task 12: A jointly-trained BERT multilingual model for offensive language detection