PingAnLifeInsurance at SemEval-2023 Task 10: Using Multi-Task Learning to Better Detect Online Sexism

Mengyuan Zhou


Abstract
This paper describes our system used in the SemEval-2023 Task 10: Towards ExplainableDetection of Online Sexism (Kirk et al., 2023). The harmful effects of sexism on the internet have impacted both men and women, yet current research lacks a fine-grained classification of sexist content. The task involves three hierarchical sub-tasks, which we addressed by employing a multitask-learning framework. To further enhance our system’s performance, we pre-trained the roberta-large (Liu et al., 2019b) and deberta-v3-large (He et al., 2021) models on two million unlabeled data, resulting in significant improvements on sub-tasks A and C. In addition, the multitask-learning approach boosted the performance of our models on subtasks A and B. Our system exhibits promising results in achieving explainable detection of online sexism, attaining a test f1-score of 0.8746 on sub-task A (ranking 1st on the leaderboard), and ranking 5th on sub-tasks B and C.
Anthology ID:
2023.semeval-1.304
Volume:
Proceedings of the 17th International Workshop on Semantic Evaluation (SemEval-2023)
Month:
July
Year:
2023
Address:
Toronto, Canada
Editors:
Atul Kr. Ojha, A. Seza Doğruöz, Giovanni Da San Martino, Harish Tayyar Madabushi, Ritesh Kumar, Elisa Sartori
Venue:
SemEval
SIG:
SIGLEX
Publisher:
Association for Computational Linguistics
Note:
Pages:
2188–2192
Language:
URL:
https://aclanthology.org/2023.semeval-1.304
DOI:
10.18653/v1/2023.semeval-1.304
Bibkey:
Cite (ACL):
Mengyuan Zhou. 2023. PingAnLifeInsurance at SemEval-2023 Task 10: Using Multi-Task Learning to Better Detect Online Sexism. In Proceedings of the 17th International Workshop on Semantic Evaluation (SemEval-2023), pages 2188–2192, Toronto, Canada. Association for Computational Linguistics.
Cite (Informal):
PingAnLifeInsurance at SemEval-2023 Task 10: Using Multi-Task Learning to Better Detect Online Sexism (Zhou, SemEval 2023)
Copy Citation:
PDF:
https://preview.aclanthology.org/naacl24-info/2023.semeval-1.304.pdf