US 12,216,684 B2Granted

Cluster-based dataset evolution

AssigneeCapital One Services, LLC

InventorsNathan Wolfe, Purva Shanker, Joshua Edwards, Gang Mei

Life of this patent

Estimating the term…

Loading patent…

Abstract

A method includes obtaining a first dataset associated with a first set of classes and a second dataset associated with a second set of classes, where the first and second datasets share features. The method also includes creating an aggregated dataset having a quality score for the second set of classes that satisfy a first threshold. Creating the aggregated dataset includes determining a first and second set of clusters based on the first and second datasets. The method includes determining a distance based on a record of the first set of clusters and a second record of the second set of clusters in a feature space. The method further includes determining a set of analysis scores based on the distance and generating an indication for a set of records associated with the set of analysis scores based on the set of analysis scores satisfying a second threshold.

Claims19

Dependent claims

A system for evolving datasets to reduce time-based labeling deterioration based on distances between records of clusters, the system comprising a computer system comprising:

one or more processors programmed with computer program instructions that, when executed, cause the computer system to perform operations comprising:
obtaining a first dataset used to train a first machine learning model configured to generate class predictions from a first set of classes, wherein each record of the first dataset is labeled with at least one class of the first set of classes, the first dataset having a first data quality score for the first set of classes that satisfies a data quality threshold;
obtaining a second dataset comprising shared features, the shared features being shared with the first dataset, wherein each record of the second dataset is labeled with at least one class of a second set of classes, at least one record of the second dataset being labeled with a new class included in the second set of classes and not included in the first set of classes;
creating, from the first dataset and the second dataset, an aggregated dataset for training a second machine learning model to be configured to generate class predictions for the second set of classes, the aggregated dataset having a second data quality score for the second set of classes that satisfy the data quality threshold, wherein creating the aggregated dataset from the first dataset and the second dataset comprises:
determining a first set of clusters of records in the first dataset, wherein each cluster of records of the first set of clusters is labeled with a respective class of the first set of classes;
determining a second set of clusters of records in the second dataset based on the second dataset, wherein each cluster of the second set of clusters is labeled with a respective class of the second set of classes;
determining a set of clustering analysis scores based on distances between records of a first cluster of the first set of clusters and records of a second cluster of the second set of clusters in a feature space of the shared features, wherein the first cluster is associated with a first class, and wherein the second cluster is associated with a second class;
and
in response to a determination that the set of clustering analysis scores satisfies a class update threshold, generating a relabeling indication for a set of records of the first dataset associated with the set of clustering analysis scores;
generating a synthesized class associated with the second class;
and
updating the set of records indicated with the relabeling indication by labeling the set of records with the synthesized class.

9 more elements

A method comprising:

obtaining, with one or more processors, a first dataset associated with a first set of classes, wherein each record of the first dataset is labeled with at least one class of the first set of classes;
obtaining, with the one or more processors, a second dataset comprising shared features that are shared with the first dataset, wherein each record of the second dataset is labeled with at least one class of a second set of classes, at least one record of the second dataset being labeled with a new class included in the second set of classes and not included in the first set of classes;
creating, with the one or more processors, an aggregated dataset based on the first dataset and the second dataset, the aggregated dataset having a data quality score for the second set of classes that satisfy a data quality threshold, wherein creating the aggregated dataset from the first dataset and the second dataset comprises:
determining a first set of clusters of records in the first dataset, wherein each cluster of records of the first set of clusters is labeled with a respective class of the first set of classes;
determining a second set of clusters of records in the second dataset based on the second dataset, wherein each cluster of the second set of clusters is labeled with a respective class of the second set of classes;
determining a distance based on a first record of a first cluster of the first set of clusters and a second record of a second cluster of the second set of clusters in a feature space of the shared features, wherein the first cluster is associated with a first class, and wherein the second cluster is associated with a second class;
determining a set of clustering analysis scores based on the distance;
and
generating, based on the set of clustering analysis scores satisfying a class update threshold, a relabeling indication for a set of records of the first dataset associated with the set of clustering analysis scores;
generating a synthesized class associated with the second class;
and
updating the set of records indicated with the relabeling indication by labeling the set of records with the synthesized class.

9 more elements

A set of non-transitory, machine-readable media storing program instructions that, when executed by a set of processors, causes the set of processors to perform operations comprising:

obtaining a first dataset associated with a first set of classes, wherein each record of the first dataset is labeled with at least one class of the first set of classes;
obtaining a second dataset comprising shared features that are shared with the first dataset, wherein each record of the second dataset is labeled with at least one class of a second set of classes, and wherein at least one record of the second dataset being labeled with a new class included in the second set of classes and not included in the first set of classes;
creating, from the first dataset and the second dataset, an aggregated dataset, the aggregated dataset having a data quality score for the second set of classes that satisfy a first threshold, wherein creating the aggregated dataset from the first dataset and the second dataset comprises:
determining a first set of clusters of records in the first dataset, wherein each cluster of records of the first set of clusters is labeled with a respective class of the first set of classes;
determining a second set of clusters of records in the second dataset based on the second dataset, wherein each cluster of the second set of clusters is labeled with a respective class of the second set of classes;
determining a distance based on a first record of a first cluster of the first set of clusters and a second record of a second cluster of the second set of clusters in a feature space of the shared features, wherein the first cluster is associated with a first class, and wherein the second cluster is associated with a second class;
determining a set of clustering analysis scores based on the distance;
and
generating, based on the set of clustering analysis scores satisfying a second threshold, an indication for a set of records of the first dataset associated with the set of clustering analysis scores;
generating a synthesized class associated with the second class;
and
updating the set of records indicated with the indication by labeling the set of records with the synthesized class.

9 more elements

Dependencies are inferred from the claim text.