US 11,893,345 B2
Inducing rich interaction structures between words for document-level event argument extraction
Amir Pouran Ben Veyseh, Eugene, OR (US); Franck Dernoncourt, San Jose, CA (US); Quan Tran, San Jose, CA (US); Varun Manjunatha, Newton, MA (US); Lidan Wang, San Jose, CA (US); Rajiv Jain, Vienna, VA (US); Doo Soon Kim, San Jose, CA (US); and Walter Chang, San Jose, CA (US)
Assigned to ADOBE, INC., San Jose, CA (US)
Filed by ADOBE INC., San Jose, CA (US)
Filed on Apr. 6, 2021, as Appl. No. 17/223,166.
Prior Publication US 2022/0318505 A1, Oct. 6, 2022
Int. Cl. G06F 40/284 (2020.01); G06F 40/211 (2020.01); G06F 40/30 (2020.01); G06N 3/08 (2023.01); G06F 40/126 (2020.01); G06N 3/044 (2023.01); G06N 3/045 (2023.01)
CPC G06F 40/284 (2020.01) [G06F 40/126 (2020.01); G06F 40/211 (2020.01); G06F 40/30 (2020.01); G06N 3/044 (2023.01); G06N 3/045 (2023.01); G06N 3/08 (2013.01)] 20 Claims
OG exemplary drawing
 
1. A method for natural language processing, comprising:
receiving a query about an event;
receiving a document comprising a plurality of words organized into a plurality of sentences, the words comprising an event trigger word corresponding to the event and an argument candidate word;
generating word representation vectors for the words by generating a first word representation vector for the event trigger word;
generating a second word representation vector for the argument candidate word;
generating a plurality of document structures including a semantic structure for the document based on the first word representation vector and the second word representation vector, a syntax structure based on a sentence-level dependency relationship, and a discourse structure representing document-level discourse information based on the plurality of sentences;
generating a relationship representation vector based on the plurality of document structures;
predicting, using a graph-based neural network, a relationship between the event trigger word and the argument candidate word based on the relationship representation vector; and
generating an answer to the query based on the predicted relationship between the event trigger word and the argument candidate word.