We propose a simple and efficient multi-hop dense retrieval approach for answering complex open-domain questions, which achieves state-of-the-art performance on two multi-hop datasets, HotpotQA and multi-evidence FEVER. Contrary to previous work, our method does not require access to any corpus-specific information, such as inter-document hyperlinks or human-annotated entity markers, and can be applied to any unstructured text corpus. Our system also yields a much better efficiency-accuracy trade-off, matching the best published accuracy on HotpotQA while being 10 times faster at inference time.
@article{arxiv.2009.12756,
title = {Answering Complex Open-Domain Questions with Multi-Hop Dense Retrieval},
author = {Wenhan Xiong and Xiang Lorraine Li and Srini Iyer and Jingfei Du and Patrick Lewis and William Yang Wang and Yashar Mehdad and Wen-tau Yih and Sebastian Riedel and Douwe Kiela and Barlas Oğuz},
journal= {arXiv preprint arXiv:2009.12756},
year = {2021}
}