Answering Complex Open-Domain Questions with Multi-Hop Dense Retrieval

Ответы на сложные вопросы в открытом домене с помощью многошагового плотного поиска
Patrick Lewis, Wen-tau Yih, Sebastian Riedel, Douwe Kiela, Jingfei Du, Yashar Mehdad, Barlas Oğuz, Srinivasan V. Iyer, Srini Iyer, W. Xiong, Xiang Lorraine Li, Wang, William Yang, Sebastian Riedel
2020-09-27

HotpotQAmulti-evidence FEVERmulti-hop dense retrievalopen-domain question answeringunstructured text corpora
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.
1
A simple and efficient multi-hop dense retrieval approach achieves state-of-the-art performance on HotpotQA and multi-evidence FEVER.
2
On HotpotQA, the system matches the best published accuracy while reducing inference time by a factor of 10, improving the efficiency-accuracy trade-off.
3
The approach applies to arbitrary unstructured text corpora rather than requiring specialized corpus structure or annotations.
4
The method answers complex open-domain questions without corpus-specific information, including inter-document hyperlinks or human-annotated entity markers.

complex open-domain question answering over unstructured text corpora

the efficiency and accuracy of multi-hop dense retrieval for answering questions requiring evidence from multiple documents

Publication Details
Publication Date
2020-09-27
Journal
Publisher
ISSN
Cited by
112
Access Type
Author Information
Authors
Patrick Lewis
Wen-tau Yih
Sebastian Riedel
Douwe Kiela
Jingfei Du
Yashar Mehdad
Barlas Oğuz
Srinivasan V. Iyer
Srini Iyer
W. Xiong
Xiang Lorraine Li
Wang, William Yang
Sebastian Riedel
Explore further
Open the scid.ai AI chat with a ready-made request: it will find papers on a similar topic and help build a literature review.
Find similar papers in the chat →
Make a presentation
100%