Knowledge graph construction method and device

Patent metadata

Field	Value
Publication number	US-11720629-B2
Application number	US-201816034799-A
Country	US
Kind code	B2
Filing date	Jul 13, 2018
Priority date	Jul 14, 2017
Publication date	Aug 8, 2023
Grant date	Aug 8, 2023

Abstract

The present invention provides a knowledge graph construction method and device. The method includes: obtaining structured data, where the structured data includes a first entity name of a first entity and attribute information corresponding to the first entity name, and the attribute information includes a first attribute and a first attribute value; performing, based on measurement of a similarity between the first entity and a second entity in a knowledge graph, entity alignment processing on the first entity, where the measurement of the similarity includes at least one of the following types: measurement of a character similarity, measurement of a structure similarity of a classification tree on which an entity is located, and measurement of an attribute similarity; and importing the structured data into the knowledge graph according to an entity alignment processing result. Embodiments may ensure correctness of data in the knowledge graph.

First claim

What is claimed is:

1. A computer-implemented knowledge graph construction method, comprising:
obtaining structured data, wherein the structured data comprises a first entity name of a first entity and attribute information corresponding to the first entity name, and the attribute information comprises a first attribute and a first attribute value;
performing, based on a measurement of similarity between the first entity and a second entity in a knowledge graph, entity alignment on the first entity, wherein the measurement of similarity comprises at least one of the following types: measurement of a character similarity, and measurement of an attribute similarity; and
importing, the structured data into the knowledge graph based on the entity alignment, wherein the importing comprises:
when the entity alignment indicates that the first entity is aligned with the second entity, and attribute alignment is performed on the first attribute of the first entity and a second attribute of the second entity, determining whether the second attribute exists in the knowledge graph;
if the second attribute does not exist in the knowledge graph, importing the first attribute and the first attribute value to the second entity; and
if the second attribute exists in the knowledge graph:
when the first attribute is a single-value attribute, determining whether the first attribute value corresponding to the first attribute conflicts with a second attribute value corresponding to the second attribute, and
if the first attribute value does not conflict with the second attribute value, performing deduplication processing;
if the first attribute value conflicts with the second attribute value, when a reliability degree of the first attribute value is higher than a reliability degree of the second attribute value, importing the first attribute value to the second entity, and deleting the second attribute value; or

when the first attribute is a multi-value attribute, and comprises a plurality of first attribute values that do not conflict with the second attribute value, determining, in the plurality of first attribute values, an attribute value different from the second attribute value, and importing the determined attribute value to the second entity.

2. The computer-implemented knowledge graph construction method according to claim 1 , wherein the performing, based on the measurement of similarity between the first entity and the second entity in the knowledge graph, entity alignment processing on the first entity comprises:
determining, according to a type of a data source of the structured data, a measurement type for performing similarity processing between the first entity and the second entity in the knowledge graph; and
performing entity alignment processing on the first entity according to the determined measurement type.

3. The computer-implemented knowledge graph construction method according to claim 2 , wherein
the performing entity alignment processing on the first entity according to the determined measurement type comprises:
determining whether a child node and a parent node of the first entity are the same as a child node and a parent node of the second entity; and if yes, determining that the entities are aligned, and if not, determining that the entities are not aligned.

4. The computer-implemented knowledge graph construction method according to claim 2 , wherein
the performing entity alignment processing on the first entity according to the determined measurement type comprises:
determining whether a character similarity between the first entity name and the second entity name in the knowledge graph is greater than a preset threshold; and if yes, determining that the entities are aligned, and if not, determining that the entities are not aligned.

5. The computer-implemented knowledge graph construction method according to claim 2 , wherein the first attribute comprises a key attribute and a non-key attribute; and
the performing entity alignment processing on the first entity according to the determined measurement type comprises:
determining whether the second attribute exists in the knowledge graph, and if yes, determining whether attribute values corresponding to the key attribute and the second attribute are the same; and if yes, determining that the entities are aligned, and if not, determining that the entities are not aligned.

6. The computer-implemented knowledge graph construction method according to claim 1 , wherein before the determining, according to a type of a data source of the structured data, a measurement type for performing similarity processing between the first entity and the second entity in the knowledge graph, the method further comprises:
obtaining a description type of each piece of attribute information; and
performing cleansing and normalization processing on each piece of attribute information according to a standard description statement corresponding to the description type, so that attribute information being semantically the same has the same description.

7. The computer-implemented knowledge graph construction method according to claim 1 , wherein the method further comprises:
in the knowledge graph, for a second attribute used to represent a relationship between entities, determining an implied relationship between entities by using a preset chain rule, and mapping the implied relationship to the knowledge graph.

8. A knowledge graph construction device, comprising a processor and a non-transitory computer-readable storage medium storing instructions that, when execute by the processor, cause the processor to perform a method comprising:
obtaining structured data, wherein the structured data comprises a first entity name of a first entity and attribute information corresponding to the first entity name, and the attribute information comprises a first attribute and a first attribute value;
performing, based on a measurement of similarity between the first entity and a second entity in a knowledge graph, entity alignment processing on the first entity, wherein the measurement of similarity comprises at least one of the following types: measurement of a character similarity, and measurement of an attribute similarity; and
importing the structured data into the knowledge graph based on the entity alignment, wherein the importing comprises:
when the entity alignment indicates that the first entity is aligned with the second entity, and attribute alignment is performed on the first attribute of the first entity and a second attribute of the second entity, determining whether the second attribute exists in the knowledge graph;
if the second attribute does not exist in the knowledge graph, importing the first attribute and the first attribute value to the second entity; and
if the second attribute exists in the knowledge graph:
when the first attribute is a single-value attribute, determining whether the first attribute value corresponding to the first attribute conflicts with a second attribute value corresponding to the second attribute, and
if the first attribute value does not conflict with the second attribute value, performing deduplication processing, or if the first attribute value conflicts with the second attribute value, when a reliability degree of the first attribute value is higher than a reliability degree of the second attribute value, importing the first attribute value to the second entity, and deleting the second attribute value; or

when the first attribute is a multi-value attribute, and comprises a plurality of first attribute values that do not conflict with the second attribute value, determining, in the plurality of first attribute values, an attribute value different from the s

Assignees

• Guangdong Shenma Search Tech Co Ltd
• Alibaba Group Holding Ltd

Inventors

• Wang Yue
• You Wei
• Zhang Weixing
• Hua Nengwei

Classifications

• G06F16/9024 Primary

  Graphs; Linked lists (G06F16/9027 takes precedence) · CPC title

• G06F18/00

  Pattern recognition · CPC title

• G06F18/22

  Matching criteria, e.g. proximity measures · CPC title

• G06F18/24323

  Tree-organised classifiers · CPC title

• G06N5/01

  Dynamic search techniques; Heuristics; Dynamic trees; Branch-and-bound · CPC title