Zero-Shot and Few-Shot Stance Detection on Varied Topics via Conditional Generation

Resources for ACL 2023 paper "Zero-Shot and Few-Shot Stance Detection on Varied Topics via Conditional Generation"

Dependencies

• python==3.9.7
• torch==1.10.1
• configargparse==1.4
• numpy==1.21.2
• transformers==4.11.3
• scikit-learn==1.0.2
• tqdm==4.62.3
• pandas=1.3.5

Training a model

In run.sh, change CUDA_DEVICE, CACHE_DIR and OUTPUT_DIR to your local setting.

Run

sh ./run.sh

Option explanations:

• wiki_path: path to the Wikipedia snippet
• predict_topic: topic prediction training
• predict_stance_neg: unlikelihood training for stance label
• predict_topic_neg: unlikelihood training for topic words (errata: we forgot to include learning topics via unlikelihood in paper description, which shares the same motivation and formulation as the one with stance label)

Reference

@inproceedings{acl-2023-stance-detection,
	title = "Zero-Shot and Few-Shot Stance Detection on Varied Topics via Conditional Generation",
	author = "Wen, Haoyang and Hauptmann, Alexander G.",
	booktitle = "Proceedings of the 61th Annual Meeting of the Association for Computational Linguistics",
	year = 2023,
}

Acknowledgement

The Wikipedia snippet is directly following the repository of the paper "Infusing Wikipedia Knowledge to Enhance Stance Detection" Zihao He, Negar Mokhberian, Kristina Lerman