Distinct Label Representations for Few-Shot Text Classification

Sora Ohashi, Junya Takayama, Tomoyuki Kajiwara, Yuki Arase


Abstract
Few-shot text classification aims to classify inputs whose label has only a few examples. Previous studies overlooked the semantic relevance between label representations. Therefore, they are easily confused by labels that are relevant. To address this problem, we propose a method that generates distinct label representations that embed information specific to each label. Our method is applicable to conventional few-shot classification models. Experimental results show that our method significantly improved the performance of few-shot text classification across models and datasets.
Anthology ID:
2021.acl-short.105
Volume:
Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on Natural Language Processing (Volume 2: Short Papers)
Month:
August
Year:
2021
Address:
Online
Venues:
ACL | IJCNLP
SIG:
Publisher:
Association for Computational Linguistics
Note:
Pages:
831–836
Language:
URL:
https://aclanthology.org/2021.acl-short.105
DOI:
10.18653/v1/2021.acl-short.105
Bibkey:
Cite (ACL):
Sora Ohashi, Junya Takayama, Tomoyuki Kajiwara, and Yuki Arase. 2021. Distinct Label Representations for Few-Shot Text Classification. In Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on Natural Language Processing (Volume 2: Short Papers), pages 831–836, Online. Association for Computational Linguistics.
Cite (Informal):
Distinct Label Representations for Few-Shot Text Classification (Ohashi et al., ACL 2021)
Copy Citation:
PDF:
https://preview.aclanthology.org/update-css-js/2021.acl-short.105.pdf
Code
 21335732529sky/difference_extractor
Data
FewRel