Filtering spurious knowledge graph relationships between labeled entities

Patent metadata

Field	Value
Publication number	US-11080491-B2
Application number	US-201916600774-A
Country	US
Kind code	B2
Filing date	Oct 14, 2019
Priority date	Oct 14, 2019
Publication date	Aug 3, 2021
Grant date	Aug 3, 2021

Abstract

Systems and techniques that facilitate spurious relationship filtration from external knowledge graphs based on distributional semantics of an input corpus are provided. In one or more embodiments, a context component can generate a context-based word embedding of one or more first terms in a document collection. The embedding can yield vector representations of the one or more first terms. The one or more first terms can correspond to knowledge terms in one or more first nodes of a knowledge graph. In one or more embodiments, a filtering component can filter out a relationship between the one or more first nodes and a second node of the knowledge graph based on a similarity value being less than a threshold. The similarity value can be a function of the vector representations of the one or more first terms. In various embodiments, cosine similarity can be used to compute the similarity value.

First claim

What is claimed is:

1. A system, comprising:
a memory that stores computer-executable components; and
a processor, operably coupled to the memory, that executes the computer-executable components stored in the memory, wherein the computer-executable components comprise:
a context component that:
generates a context-based embedding of one or more first terms in a document collection, thereby yielding vector representations of the one or more first terms, wherein the one or more first terms correspond to knowledge terms in one or more first nodes of a knowledge graph; and
determines that at least two of the one or more first terms that have vector representations that have a similarity value that meets a defined criterion have a hypernymy relation between them, and determines that at least two of the one or more first terms that have vector representations that fail to have a similarity value that meets the defined criterion do not have a hypernymy relation between them, wherein the vector representations are compared to determine whether the hypernymy relation from the knowledge graph for the at least two of the one or more first terms is spurious; and

a filtering component that filters out a relationship between the one or more first nodes and a second node of the knowledge graph based on the similarity value.

2. The system of claim 1 , wherein:
the similarity value is based on a cosine similarity between the vector representations of the one or more first terms and a vector representation of a second term in the document collection corresponding to the second node.

3. The system of claim 1 , wherein:
the similarity value is based on average pairwise cosine similarities between the vector representations of the one or more first terms.

4. The system of claim 1 , wherein:
the similarity value is based on cosine similarities between the vector representations of the one or more first terms and a vector representation of a prototypical term in the one or more first terms.

5. The system of claim 1 , wherein the knowledge terms are part of one or more labels, descriptions, definitions, or other text associated with the one or more first nodes.

6. The system of claim 1 , wherein the relationship is one from the group consisting of a hypernym-hyponym relation, a synonymy relation, an antonymy relation, an entailment relation, and a partonomy relation.

7. The system of claim 1 , wherein the one or more first terms correspond to the knowledge terms lexically, orthographically, morphologically, syntactically, or semantically.

8. The system of claim 1 , wherein the context component generates the embedding of the one or more first terms via a neural network that employs a Continuous Bag of Words or Skip Gram methodology.

9. A computer-implemented method, comprising:
generating, by a device operatively coupled to a processor, a context-based embedding of one or more first terms in a document collection, thereby yielding vector representations of the one or more first terms, wherein the one or more first terms correspond to knowledge terms in one or more first nodes of a knowledge graph;
determining, by the device, that at least two of the one or more first terms that have vector representations that have a similarity value that meets a defined criterion have a hypernymy relation between them, and determine that at least two of the one or more first terms that have vector representations that fail to have a similarity value that meets the defined criterion do not have a hypernymy relation between them, wherein the vector representations are compared to determine whether the hypernymy relation from the knowledge graph for the at least two of the one or more first terms is spurious; and
filtering out, by the device, a relationship between the one or more first nodes and a second node of the knowledge graph based on the similarity value.

10. The computer-implemented method of claim 9 , wherein:
the similarity value is based on a cosine similarity between the vector representations of the one or more first terms and a vector representation of a second term in the document collection corresponding to the second node.

11. The computer-implemented method of claim 9 , wherein:
the similarity value is based on average pairwise cosine similarities between the vector representations of the one or more first terms.

12. The computer-implemented method of claim 9 , wherein:
the similarity value is based on cosine similarities between the vector representations of the one or more first terms and a vector representation of a prototypical term in the one or more first terms.

13. The computer-implemented method of claim 9 , wherein the knowledge terms are part of one or more labels, descriptions, definitions, or other text associated with the one or more first nodes.

14. The computer-implemented method of claim 9 , wherein the relationship is one from the group consisting of a hypernym-hyponym relation, a synonymy relation, an antonymy relation, an entailment relation, and a partonomy relation.

15. The computer-implemented method of claim 9 , wherein the one or more first terms correspond to the knowledge terms lexically, orthographically, morphologically, syntactically, or semantically.

16. The computer-implemented method of claim 9 , wherein the generating the embedding of the one or more first terms is performed with a neural network that employs a Continuous Bag of Words or Skip Gram methodology.

17. A computer program product for facilitating spurious relationship filtration, the computer program product comprising a computer readable storage medium having program instructions embodied therewith, the program instructions executable by a processing component to cause the processing component to:
generate a context-based embedding of one or more first terms in a document collection, thereby yielding vector representations of the one or more first terms, wherein the one or more first terms correspond to knowledge terms in one or more first nodes of a knowledge graph;
determine that at least two of the one or more first terms that have vector representations that have a similarity value that meets a defined criterion have a supplier relation between them, and determine that at least two of the one or more first terms that have vector representations that fail to have a similarity value that meets the defined criterion do not have a supplier relation between them, wherein the vector representations are compared to determine whether the supplier relation from the knowledge graph for the at least two of the one or more first terms is spurious; and
filter out a relationship between the one or more first nodes and a second node of the knowledge graph based on the similarity value.

18. The computer program product of claim 17 , wherein:
the similarity value is based on a cosine similarity between the vector representations of the one or more first terms and a vector representation of a second term in the document collection corresponding to the second node.

19. The computer program product of claim 17 , wherein:
the similarity value is based on average pairwise cosine similarities between the vector representations of the one or more first terms.

20. The computer program product of claim 17 , wherein:
the similarity value is based on cosine similarities between the vector representations of the one or more first terms and a vector representation of a prototypical term in the one or more first terms.

Assignees

• IBM

Inventors

• Mihindukulasooriya Nandana
• Farrell Robert G
• Fauceglia Nicolas Rodolfo
• Gliozzo Alfio Massimiliano

Classifications

• G06F16/335

  Filtering based on additional data, e.g. user or group profiles (filtering in web context G06F16/9535, G06F16/9536) · CPC title

• G06F16/9024

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

• G06F40/295

  Named entity recognition · CPC title

• G06F40/268

  Morphological analysis · CPC title

• G06F40/30 Primary

  Semantic analysis · CPC title