US 12,013,902 B2
Inter-document attention mechanism
Chenyan Xiong, Bellevue, WA (US); Chen Zhao, Greenbelt, MD (US); Corbin Louis Rosset, Seattle, WA (US); Paul Nathan Bennett, Redmond, WA (US); Xia Song, Redmond, WA (US); and Saurabh Kumar Tiwary, Bellevue, WA (US)
Assigned to Microsoft Technology Licensing, LLC, Redmond, WA (US)
Filed by Microsoft Technology Licensing, LLC, Redmond, WA (US)
Filed on Jul. 18, 2022, as Appl. No. 17/866,854.
Application 17/866,854 is a continuation of application No. 16/582,582, filed on Sep. 25, 2019, granted, now 11,423,093.
Prior Publication US 2022/0374479 A1, Nov. 24, 2022
This patent is subject to a terminal disclaimer.
Int. Cl. G06F 16/93 (2019.01); G06F 16/903 (2019.01); G06N 3/04 (2023.01); G06N 3/08 (2023.01)
CPC G06F 16/93 (2019.01) [G06F 16/90335 (2019.01); G06N 3/04 (2013.01); G06N 3/08 (2013.01)] 20 Claims
OG exemplary drawing
 
1. A method performed on a computing device, the method comprising:
receiving a query;
retrieving result documents related to the query;
establishing an evidence graph having nodes representing the result documents and links connecting individual result documents;
obtaining contextualized semantic representations for individual words in a second result document by propagating inter-document attention to the individual words from a first result document that is linked to the second result document in the evidence graph, the propagating being based at least on respective similarities of the individual words in the second result document to an entirety of the first result document;
processing the contextualized semantic representations of the individual words in the second result document to obtain an answer to the query; and
outputting the answer in response to the query.