AlignRE

This repository contains the official implementation for the paper: AlignRE: An Encoding and Semantic Alignment Approach for Zero-Shot Relation Extraction. The paper has been accepted to appear at ACL 2024 Findings.

Dataset

• FewRel (Xu et al., 2018)
• WikiZSL (Daniil Sorokin and Iryna Gurevych, 2017)

Place them to the /data folder.

Requirements

The main requirements are:

• python==3.9.7
• pytorch==1.13.1
• transformers==4.44.2
• sentence-transformers==2.2.2
• numpy==1.22.3

Run

1. Download the pretrained Bert weights to folder './BERT_MODELS/bert-base-uncased'.
2. Download the pretrained Sentence-Bert weights to folder './BERT_MODELS/stsb-bert-base'.
3. Set hyperparameters and run train.py

Cite

@inproceedings{li2024alignre,
  title={AlignRE: An Encoding and Semantic Alignment Approach for Zero-Shot Relation Extraction},
  author={Li, Zehan and Zhang, Fu and Cheng, Jingwei},
  booktitle={Findings of the Association for Computational Linguistics ACL 2024},
  pages={2957--2966},
  year={2024}
}

Acknowledgement

The framework of AlignRE is based on ZS-BERT and RE-Matching. Their contributions have greatly helped in the development and implementation of our code. We appreciate their efforts and the open-source community for fostering collaboration and knowledge sharing.