Knowledge Stimulated Contrastive Prompting for Low-Resource Stance Detection

Kai Zheng, Qingfeng Sun, Yaming Yang, Fei Xu


Abstract
Stance Detection Task (SDT) aims at identifying the stance of the sentence towards a specific target and is usually modeled as a classification problem. Backgound knowledge is often necessary for stance detection with respect to a specific target, especially when there is no target explicitly mentioned in text. This paper focuses on the knowledge stimulation for low-resource stance detection tasks. We firstly explore to formalize stance detection as a prompt based contrastive learning task. At the same time, to make prompt learning suit to stance detection, we design a template mechanism to incorporate corresponding target into instance representation. Furthermore, we propose a masked language prompt joint contrastive learning approach to stimulate the knowledge inherit from the pre-trained model. The experimental results on three benchmarks show that knowledge stimulation is effective in stance detection accompanied with our proposed mechanism.
Anthology ID:
2022.findings-emnlp.83
Volume:
Findings of the Association for Computational Linguistics: EMNLP 2022
Month:
December
Year:
2022
Address:
Abu Dhabi, United Arab Emirates
Editors:
Yoav Goldberg, Zornitsa Kozareva, Yue Zhang
Venue:
Findings
SIG:
Publisher:
Association for Computational Linguistics
Note:
Pages:
1168–1178
Language:
URL:
https://aclanthology.org/2022.findings-emnlp.83
DOI:
10.18653/v1/2022.findings-emnlp.83
Bibkey:
Cite (ACL):
Kai Zheng, Qingfeng Sun, Yaming Yang, and Fei Xu. 2022. Knowledge Stimulated Contrastive Prompting for Low-Resource Stance Detection. In Findings of the Association for Computational Linguistics: EMNLP 2022, pages 1168–1178, Abu Dhabi, United Arab Emirates. Association for Computational Linguistics.
Cite (Informal):
Knowledge Stimulated Contrastive Prompting for Low-Resource Stance Detection (Zheng et al., Findings 2022)
Copy Citation:
PDF:
https://preview.aclanthology.org/improve-issue-templates/2022.findings-emnlp.83.pdf