Splitting incorrectly resolved entities using minimum cut

A computer-implemented method comprising:

• obtaining a dataset indication of a dataset with data records each resolved to describe a single entity in an entity resolution, wherein the data records comprise multiple peripheral records resolved to describe the single entity in the entity resolution based on matching data of a central record of the data records, wherein the central record comprises first data matching data of a first peripheral record of the multiple peripheral records, and wherein the central record comprises second data matching data of a second peripheral record of the multiple peripheral records;
• determining that at least two data records of the dataset do not describe a same entity;
• based on the determining that the at least two data records of the dataset do not describe the same entity, generating a record indication indicating that at least one of the first peripheral record and the second peripheral record does not describe the single entity by:
• setting a first one of the data records as a source;
• setting a second one of the data records as a sink;
• generating a data structure identifying record linkage information for records of the dataset, wherein the record linkage information indicates:
• linked records that have matching data, and
• one or more pathways between the source and the sink along the linked records;
• identifying a first set of linked records that when unlinked would disconnect all of the one or more pathways between the source and the sink;
• identifying a second set of linked records that when unlinked would disconnect all of the one or more pathways between the source and the sink, wherein the first set of linked records is different than the second set of linked records;
• generating a respective metric for each of respective linked records of the first set of linked records and the second set of linked records;
• computing a first sum of one or more metrics corresponding to linked records of the first set;
• computing a second sum of one or more metrics corresponding to linked records of the second set;
• identifying one or more connections of the one or more pathways to unlink based on the first sum and the second sum;
• and
• outputting the record indication that at least one of the first peripheral record and the second peripheral record does not describe the single entity.

A computing device comprising processor and memory, the memory containing instructions executable by the processor wherein the computing device is configured to:

• obtain a dataset indication of a dataset with data records each resolved to describe a single entity in an entity resolution, wherein the data records comprise multiple peripheral records resolved to describe the single entity in the entity resolution based on matching data of a central record of the data records, wherein the central record comprises first data matching data of a first peripheral record of the multiple peripheral records, and wherein the central record comprises second data matching data of a second peripheral record of the multiple peripheral records;
• determine that at least two data records of the dataset do not describe a same entity;
• based on determining that the at least two data records of the dataset do not describe the same entity, generate a record indication indicating that at least one of the first peripheral record and the second peripheral record does not describe the single entity by:
• setting a first one of the data records as a source;
• setting a second one of the data records as a sink;
• generating a data structure identifying record linkage information for records of the dataset, wherein the record linkage information indicates:
• linked records that have matching data, and
• one or more pathways between the source and the sink along the linked records;
• identifying a first set of linked records that when unlinked would disconnect all of the one or more pathways between the source and the sink;
• identifying a second set of linked records that when unlinked would disconnect all of the one or more pathways between the source and the sink, wherein the first set of linked records is different than the second set of linked records;
• generating a respective metric for each of respective linked records of the first set of linked records and the second set of linked records;
• computing a first sum of one or more metrics corresponding to linked records of the first set;
• computing a second sum of one or more metrics corresponding to linked records of the second set;
• identifying one or more connections of the one or more pathways to unlink based on the first sum and the second sum;
• and
• output the record indication that at least one of the first peripheral record and the second peripheral record does not describe the single entity.