This repository includes the original implementation of our ACL2022 paper "Answering Open-Domain Multi-Answer Questions via a Recall-then-Verify Framework", which contains the following two branches:
- retriever: code for finetuning a retriever
- rectify: code for training a recall-then-verify system
Download links for data and checkpoints:
- All source data and inferred results: https://cloud.tsinghua.edu.cn/d/491c750515b94c73bbbb/
- Finetuned retrievers and inferred passage embeddings: https://cloud.tsinghua.edu.cn/d/85c358be556e4ce7b8ec/
- Recallers and verifiers: https://cloud.tsinghua.edu.cn/d/71a2177eefb34a33848c/
Due to the limited upload size, zip files have been chunked into smaller ones whose names share the same prefix except the last letter (e.g., data.zipchunkaa and data.zipchunkab are the first and the second chunk of the original zip file data.zip, respectively). Please merge chunks with the same prefix before unzipping them (e.g., cat data.zipchunka* > data.zip && unzip data.zip).
If you find this work useful, please cite our paper:
@inproceedings{DBLP:conf/acl/ShaoH22,
author = {Zhihong Shao and
Minlie Huang},
editor = {Smaranda Muresan and
Preslav Nakov and
Aline Villavicencio},
title = {Answering Open-Domain Multi-Answer Questions via a Recall-then-Verify
Framework},
booktitle = {Proceedings of the 60th Annual Meeting of the Association for Computational
Linguistics (Volume 1: Long Papers), {ACL} 2022, Dublin, Ireland,
May 22-27, 2022},
pages = {1825--1838},
publisher = {Association for Computational Linguistics},
year = {2022},
url = {https://aclanthology.org/2022.acl-long.128},
timestamp = {Wed, 18 May 2022 15:21:43 +0200},
biburl = {https://dblp.org/rec/conf/acl/ShaoH22.bib},
bibsource = {dblp computer science bibliography, https://dblp.org}
}