Learning to retrieve reasoning paths from the Wikipedia graph

Learning to retrieve reasoning paths from the Wikipedia graph

We develop a graph-based trainable retriever-reader framework which sequentially retrieves evidence paragraphs (reasoning paths) from the entire English Wikipedia to answer open-domain questions. Our framework shows state-of-the-art performance on HotpotQA, SQuAD Open, and Natural Questions Open without any architectural changes, showcasing its effectiveness and robustness for open-domain question answering. Background“

6 mentions: @AkariAsai@storyneedle
Date: 2020/02/24 17:00

Referring Tweets

@AkariAsai Our #ICLR2020 camera-ready version, code, and blog are now available! paper: t.co/VC1FxuGBMI code: t.co/rYUH3QYMdG blog: t.co/EpeFmpufdR You can train, evaluate, and run an interactive demo on your machine. We also release the models for reproducibility. t.co/nz2L0U2rfp
@storyneedle I mentioned recently the need for graph databases to work with real content, not just entities and enumerated values. This work by Salesforce shows the direction that's needed: open domain graphs looking at paragraph level info t.co/ZOAlkQ4J4z

Related Entries

Read more [2003.00381] Statistical power for cluster analysiscontact arXivarXiv Twitter
0 users, 6 mentions 2020/03/03 23:20
Read more [2004.10240] Neural forecasting: Introduction and literature overviewopen searchopen navigation menu...
0 users, 6 mentions 2020/04/23 21:51
Read more A 2020 Vision of Linear Algebra | MIT OpenCourseWare
0 users, 12 mentions 2020/05/09 12:52
Read more [2006.11477] wav2vec 2.0: A Framework for Self-Supervised Learning of Speech Representationsopen sea...
0 users, 6 mentions 2020/06/23 14:21
Read more Knowledge Graphs in Natural Language Processing @ ACL 2020 | by Michael Galkin | Jul, 2020 | Towards...
1 users, 7 mentions 2020/07/11 21:49