MIPT-NSU-UTMN at SemEval-2021 Task 5: Ensembling Learning with Pre-trained Language Models for Toxic Spans Detection

Mikhail Kotyushev, Anna Glazkova, Dmitry Morozov


Abstract
This paper describes our system for SemEval-2021 Task 5 on Toxic Spans Detection. We developed ensemble models using BERT-based neural architectures and post-processing to combine tokens into spans. We evaluated several pre-trained language models using various ensemble techniques for toxic span identification and achieved sizable improvements over our baseline fine-tuned BERT models. Finally, our system obtained a F1-score of 67.55% on test data.
Anthology ID:
2021.semeval-1.124
Volume:
Proceedings of the 15th International Workshop on Semantic Evaluation (SemEval-2021)
Month:
August
Year:
2021
Address:
Online
Editors:
Alexis Palmer, Nathan Schneider, Natalie Schluter, Guy Emerson, Aurelie Herbelot, Xiaodan Zhu
Venue:
SemEval
SIG:
SIGLEX
Publisher:
Association for Computational Linguistics
Note:
Pages:
913–918
Language:
URL:
https://aclanthology.org/2021.semeval-1.124
DOI:
10.18653/v1/2021.semeval-1.124
Bibkey:
Cite (ACL):
Mikhail Kotyushev, Anna Glazkova, and Dmitry Morozov. 2021. MIPT-NSU-UTMN at SemEval-2021 Task 5: Ensembling Learning with Pre-trained Language Models for Toxic Spans Detection. In Proceedings of the 15th International Workshop on Semantic Evaluation (SemEval-2021), pages 913–918, Online. Association for Computational Linguistics.
Cite (Informal):
MIPT-NSU-UTMN at SemEval-2021 Task 5: Ensembling Learning with Pre-trained Language Models for Toxic Spans Detection (Kotyushev et al., SemEval 2021)
Copy Citation:
PDF:
https://aclanthology.org/2021.semeval-1.124.pdf
Code
 morozowdmitry/semeval21