Abstract: One or more trained embedding generation artificial intelligence models are executed to generate a plurality of record attribute embeddings. The plurality of record attribute embeddings represents a plurality of attributes of data of a plurality of records. Grouping of the plurality of record attribute embeddings is performed. The grouping of a record attribute embedding includes grouping attribute values of the record attribute embedding into one or more groups of attribute values. The performing grouping provides a plurality of groups of attribute values for the plurality of record attribute embeddings. Selected records are compared to provide a set of matched records. The comparing, based on a group of attribute values, includes comparing records that include one or more attribute values grouped in the group of attribute values providing a subset of matched records of the set of matched records. The set of matched records is stored in an accessible computer location.

Abstract: Described are techniques for disintegrating an entity into smaller entities. A graph (“first graph”) for the entity of records to be disintegrated is constructed, where each vertex of the first graph represents a record in the entity of records to be disintegrated. The edges in the first graph connecting records in the entity of records represent matching links between the records, where each edge is associated with a weight corresponding to a similarity score. Furthermore, two or more additional graphs representing two or more sub-entities of the entity of records to be disintegrated are constructed. Such graphs are constructed based on selecting edges with a maximum weight out of the edges connected between each pair of records in the first graph or based on the number of connections each record has with other records in the first graph exceeding a threshold value.

Abstract: Described are techniques for disintegrating an entity into smaller entities. A graph (“first graph”) for the entity of records to be disintegrated is constructed, where each vertex of the first graph represents a record in the entity of records to be disintegrated. The edges in the first graph connecting records in the entity of records represent matching links between the records, where each edge is associated with a weight corresponding to a similarity score. Furthermore, two or more additional graphs representing two or more sub-entities of the entity of records to be disintegrated are constructed. Such graphs are constructed based on selecting edges with a maximum weight out of the edges connected between each pair of records in the first graph or based on the number of connections each record has with other records in the first graph exceeding a threshold value.

Abstract: One or more trained embedding generation artificial intelligence models are executed to generate a plurality of record attribute embeddings. The plurality of record attribute embeddings represents a plurality of attributes of data of a plurality of records. Grouping of the plurality of record attribute embeddings is performed. The grouping of a record attribute embedding includes grouping attribute values of the record attribute embedding into one or more groups of attribute values. The performing grouping provides a plurality of groups of attribute values for the plurality of record attribute embeddings. Selected records are compared to provide a set of matched records. The comparing, based on a group of attribute values, includes comparing records that include one or more attribute values grouped in the group of attribute values providing a subset of matched records of the set of matched records. The set of matched records is stored in an accessible computer location.

Abstract: A method, system, and computer program product are configured to create a tuned data record matching model by adjusting values of one or more parameters in a data record matching model based on a second training data set labeled at a data record level, wherein the data record matching model is initially trained using a first training data set labeled at an attribute level.

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.

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.

Abstract: Records linking is provided. Two records are selected from a plurality of records corresponding to a customer for pair-wise record comparison. It is determined whether the two records are included in different entities. A local auto-link-threshold value of the different entities is identified in response to determining that the two records are included in different entities. An attribute comparison is performed between the two records. A comparison score is generated based on the attribute comparison between the two records. It is determined whether the comparison score is greater than the local auto-link-threshold value of the different entities. The two records are linked in response to determining that the comparison score is greater than the local auto-link-threshold value of the different entities.

Abstract: A multilevel clustered data set for multidimensional vectors is created by defining a plurality of clusters based on each of the signed dimensions of the vectors, each dimension functioning as an axis. Vectors are assigned to each cluster by measuring cosine similarity between a vector and each axis. Sub-clusters are defined as ranges of cosine similarity values within a cluster, and each vector is assigned into the appropriate range based on their cosine similarity value with the axis of the cluster. Searching for a matching vector to a new vector is efficiently achieved in near-constant time by measuring cosine similarity for the new vector with each axis to identify the closest cluster, reusing the cosine similarity of the new vector and axis to determine which sub-cluster corresponds to the appropriate range of values, and then comparing each vector within the sub-cluster until a match is found or ruled out.

Abstract: A multilevel clustered data set for multidimensional vectors is created by defining a plurality of clusters based on each of the signed dimensions of the vectors, each dimension functioning as an axis. Vectors are assigned to each cluster by measuring cosine similarity between a vector and each axis. Sub-clusters are defined as ranges of cosine similarity values within a cluster, and each vector is assigned into the appropriate range based on their cosine similarity value with the axis of the cluster. Searching for a matching vector to a new vector is efficiently achieved in near-constant time by measuring cosine similarity for the new vector with each axis to identify the closest cluster, reusing the cosine similarity of the new vector and axis to determine which sub-cluster corresponds to the appropriate range of values, and then comparing each vector within the sub-cluster until a match is found or ruled out.