US 11,783,131 B2
Knowledge graph fusion
Zhong Fang Yuan, Xian (CN); Chen Gao, Xian (CN); Tong Liu, Xian (CN); De Shuo Kong, Beijing (CN); Ci-Wei Lan, Keelung (TW); and Rong Fu He, Beijing (CN)
Assigned to International Business Machines Corporation, Armonk, NY (US)
Filed by International Business Machines Corporation, Armonk, NY (US)
Filed on Sep. 10, 2020, as Appl. No. 17/17,258.
Prior Publication US 2022/0075948 A1, Mar. 10, 2022
Int. Cl. G06F 40/295 (2020.01); G06F 16/901 (2019.01); G06N 5/02 (2023.01); G06N 20/00 (2019.01)
CPC G06F 40/295 (2020.01) [G06F 16/9024 (2019.01); G06N 5/02 (2013.01); G06N 20/00 (2019.01)] 13 Claims
OG exemplary drawing
 
1. A computer-implemented method comprising:
extracting contextual information from a plurality of documents;
generating, based on the extracted contextual information, a knowledge graph for each document of the plurality of documents;
analyzing each knowledge graph to determine if one or more entities of each knowledge graph are linked, wherein the analyzing comprises:
determining a set of textual mentions related to each of the one or more entities of each knowledge graph; and
mapping one or more textual mentions from the set of textual mentions related to each of the one or more entities to a same or a similar entity in a different knowledge graph based on a similarity score, wherein the similarity score is weighted based, in part, on one or more reliability characteristics of a source of each document of the plurality of documents;
fusing, in response to an entity in a first knowledge graph being linked to an entity in a second knowledge graph, the first knowledge graph with the second knowledge graph to create a fused knowledge graph;
receiving a first natural language query (NLQ) from a user regarding the one or more entities;
outputting, using the fused knowledge graph, an answer to the first NLQ;
receiving a second NLQ from the same user, wherein the second NLQ is substantially similar to the first NLQ;
determining, based on the second NLQ being substantially similar to the first NLQ, that an accuracy value of the answer to the first NLQ is low;
analyzing, using machine learning, the accuracy value of the answer to the first NLQ that was provided using the fused knowledge graph; and
modifying, using machine learning and based on the low accuracy value for the answer, algorithms for traversing the fused knowledge graph to obtain a higher accuracy value when providing a replacement answer.