Entity explanation in data management

Patent metadata

Field	Value
Publication number	US-12045291-B2
Application number	US-202217980477-A
Country	US
Kind code	B2
Filing date	Nov 3, 2022
Priority date	Nov 3, 2022
Publication date	Jul 23, 2024
Grant date	Jul 23, 2024

Abstract

Records can be matched by a graph neural network model performing entity resolution on the records, and representing each record as a respective node in a graph. Record matching explanations can be generated, each record matching explanation indicating a first set of attributes, and a first set of corresponding values, used for the matching at least two of the records. Nodes can be clustered into a plurality of clusters by aggregating the record matching explanations and, based on the record matching explanations, determining which of the records have high importance values, in the first set of values, that match. At least one cluster explanation can be generated, the cluster explanation indicating a second set of attributes, and a second set of values corresponding to the second set of attributes, used for the clustering the nodes. The record matching explanation and the cluster explanation can be output.

First claim

What is claimed is:

1. A method, comprising:
matching a plurality of records by a graph neural network model performing entity resolution on the plurality of records, and representing each of the plurality of records as a respective node in a graph;
generating a plurality of record matching explanations, each record matching explanation indicating a first set of attributes, and a first set of values corresponding to the first set of attributes, used for the matching at least two of the plurality of records;
clustering the nodes into a plurality of clusters by aggregating the plurality of record matching explanations and, based on the plurality of record matching explanations, determining which of the plurality of records have high importance values, in the first set of values, that match;
generating, using a processor, at least one cluster explanation, the cluster explanation indicating a second set of attributes, and a second set of values corresponding to the second set of attributes, used for the clustering the nodes;
outputting the record matching explanation and the cluster explanation; and
training the graph neural network model to perform the entity resolution, the training comprising using probabilistic matching engine output data as input data to the graph neural network.

2. The method of claim 1 , further comprising:
identifying at least one anomalous node or cluster in the graph;
outputting data indicating the at least one anomalous node or cluster in the graph;
receiving a user input indicating the at least one anomalous node or cluster to unlink from the graph;
responsive to the receiving the user input indicating the at least one anomalous node or cluster to unlink from the graph, determining an impact on the graph of unlinking the at least one anomalous node or cluster from the graph by analyzing the graph without the indicated anomalous nodes and/or anomalous clusters to unlink from the graph; and
outputting the determined impact on the graph of unlinking the indicated anomalous nodes and/or anomalous clusters from the graph.

3. The method of claim 1 , wherein:
the generating the plurality of record matching explanations comprises assigning, to each value in the first set of values, a score indicating how significant the value is toward a determination that a pair of records match; and
the determining which of the plurality of records have high importance values comprises determining values in the first set of values having highest scores.

4. The method of claim 3 , wherein the cluster explanation indicates, for each value in the second set of values, the score indicating how significant the value is toward the determination that the pair of records match.

5. The method of claim 1 , further comprising:
assigning a respective score to each node of the graph based on analyzing the record represented by the node; and
determining a representative record for the entity based on the scores assigned to the nodes.

6. The method of claim 1 , further comprising:
generating an inter-cluster edge explanation, the inter-cluster edge explanation indicating at least one dissimilarity between at least two nodes.

7. A system, comprising:
a processor programmed to initiate executable operations comprising:
matching a plurality of records by a graph neural network model performing entity resolution on the plurality of records, and representing each of the plurality of records as a respective node in a graph;
generating a plurality of record matching explanations, each record matching explanation indicating a first set of attributes, and a first set of values corresponding to the first set of attributes, used for the matching at least two of the plurality of records;
clustering the nodes into a plurality of clusters by aggregating the plurality of record matching explanations and, based on the plurality of record matching explanations, determining which of the plurality of records have high importance values, in the first set of values, that match;
generating at least one cluster explanation, the cluster explanation indicating a second set of attributes, and a second set of values corresponding to the second set of attributes, used for the clustering the nodes;
outputting the record matching explanation and the cluster explanation; and

training the graph neural network model to perform the entity resolution, the training comprising using probabilistic matching engine output data as input data to the graph neural network.

8. The system of claim 7 , the executable operations further comprising:
identifying at least one anomalous node or cluster in the graph;
outputting data indicating the at least one anomalous node or cluster in the graph;
receiving a user input indicating the at least one anomalous node or cluster to unlink from the graph;
responsive to the receiving the user input indicating the at least one anomalous node or cluster to unlink from the graph, determining an impact on the graph of unlinking the at least one anomalous node or cluster from the graph by analyzing the graph without the indicated anomalous nodes and/or anomalous clusters to unlink from the graph; and
outputting the determined impact on the graph of unlinking the indicated anomalous nodes and/or anomalous clusters from the graph.

9. The system of claim 7 , wherein:
the generating the plurality of record matching explanations comprises assigning, to each value in the first set of values, a score indicating how significant the value is toward a determination that a pair of records match; and
the determining which of the plurality of records have high importance values comprises determining values in the first set of values having highest scores.

10. The system of claim 9 , wherein the cluster explanation indicates, for each value in the second set of values, the score indicating how significant the value is toward the determination that the pair of records match.

11. The system of claim 7 , the executable operations further comprising:
assigning a respective score to each node of the graph based on analyzing the record represented by the node; and
determining a representative record for the entity based on the scores assigned to the nodes.

12. The system of claim 7 , the executable operations further comprising:
generating an inter-cluster edge explanation, the inter-cluster edge explanation indicating at least one dissimilarity between at least two nodes.

13. A computer program product, comprising:
one or more computer readable storage mediums having program code stored thereon, the program code stored on the one or more computer readable storage mediums collectively executable by a data processing system to initiate operations including:
matching a plurality of records by a graph neural network model performing entity resolution on the plurality of records, and representing each of the plurality of records as a respective node in a graph;
generating a plurality of record matching explanations, each record matching explanation indicating a first set of attributes, and a first set of values corresponding to the first set of attributes, used for the matching at least two of the plurality of records;
clustering the nodes into a plurality of clusters by aggregating the plurality of record matching explanations and, based on the plurality of record matching explanations, determining which of the plurality of records have high importance values, in the first set of values, that match;
generating at least one cluster explanation, the cluster explanation indicating a second set of attributes, and a second set of values corresponding to

Assignees

• IBM

Inventors

• Ameen Muhammed Abdul Majeed
• Ganesan Balaji
• Saha Avirup
• Seth Abhishek
• Sharma Devbrat
• Agarwal Arvind
• Naganna Soma Shekar
• Mehta Sameep

Classifications

• G06N7/01

  Probabilistic graphical models, e.g. probabilistic networks · CPC title

• G06N3/045

  Combinations of networks · CPC title

• G06N20/00

  Machine learning · CPC title

• G06F16/906 Primary

  Clustering; Classification · CPC title