Multi-level conflict-free entity clusterings
AssigneeAmperity, Inc.
InventorsYan Yan, Stephen Keith Meyles, Graeme Andrew Kyle Roche, Jeffrey Allen Stokes, Carlos Minoru Sakoda, Dan Suciu
Life of this patent
Estimating the term…
Abstract
The present disclosure relates clustering similar data records together in a hierarchical clustering scheme. Each tier in a cluster corresponds to a minimal match score, which reflects a degree of confidence. In this respect, a higher confidence may lead to smaller sized clusters while a lower confidence may lead to larger sized clusters. Ordinal classification may be used to generate hierarchical clusters. In some embodiments, hierarchical clustering with conflict resolution is used to resolve user-defined hard conflicts in each tier of the clustering results.
Claims20
A system comprising:
- a database that stores at least one database table comprising a plurality of records;
- and
- a memory coupled to a processor, the memory comprising a plurality of instructions that cause the processor to:
- identify a subset of records among the plurality of records according to at least one blocking operation;
- perform a plurality of classifications, by an ordinal classifier configured by ordinal training data that comprises a set of non-binary, ordinal labels, among the subset of records to generate corresponding match scores and one or more indications of a hard conflict;
- determine to not merge at least one record among the subset of records with a cluster of a plurality of records among the subset of records when a) at least one match score indicates a sufficiently strong match and b) a determination that a hard conflict exists between the at least one record and a plurality of records within the cluster;
- and
- generate a hierarchical clustering for the subset of records, the hierarchical clustering comprising a plurality of tiers, wherein each tier corresponds to a respective threshold match score, wherein a first tier among the plurality of tiers encompasses a second tier among the plurality of tiers.
(canceled)
A computer implemented method comprising:
- performing a first set of ordinal classifications, by an ordinal classifier configured by ordinal training data that comprises a set of ordinal labels, on a subset of records derived from a plurality of records contained within at least one database table to generate a first set of corresponding match scores, wherein the ordinal labels are non-binary;
- merging at least two records within the subset of records to generate an initial cluster within the subset of records, according to the first set of corresponding match scores;
- performing a second set of ordinal classifications, by the ordinal classifier, among the subset of records, including the initial cluster, to generate a subsequent set of corresponding match scores, and
- generating a hierarchical clustering for the subset of records, the hierarchical clustering comprising a plurality of tiers, each tier corresponding to a respective ordinal label among the set of ordinal labels, wherein each tier corresponds to a respective threshold match score.
One or more non-transitory computer-readable, non-volatile storage memory comprising stored instructions that are executable and, responsive to execution by a computing device, the computing device performs operations comprising:
- performing a first set of ordinal classifications, by an ordinal classifier configured by ordinal training data that comprises a set of ordinal labels, among a subset of records of at least one database table to generate a first set of corresponding match scores, wherein the ordinal labels are non-binary;
- merging at least two records within the subset of records to generate an initial cluster within the subset of records, according to the first set of corresponding match scores;
- performing at a second set of ordinal classifications, by the ordinal classifier, among the subset of records, including the initial cluster, to generate a subsequent set of corresponding match scores, and
- generating a hierarchical clustering for the subset of records, the hierarchical clustering comprising a plurality of tiers, wherein each tier corresponds to a respective ordinal label among the set of ordinal labels.
Dependencies are inferred from the claim text.