Hybrid model for data record classification
AssigneePlatts (U.K.) Limited
InventorsVikas Mehrotra, Thomas Goatly, Minesh Soni
Life of this patent
Estimating the term…
Abstract
A method, apparatus, system, and computer program code for classifying data. A number of virtual datasets is generated at multiple granularities across a number of data dimensions. Data records that correspond to a combination of data dimensions are identified in a virtual dataset. A number of defined rules are applied to: determine a set of granularities for each of the number data dimensions for comparing the data records that were identified; and determine a model for comparing the data records that were identified. The data records are classified according to the model and the set of granularities that were determined for the number data dimensions.
Claims21
A method for classifying data records, the method comprising:
- generating, by a number of processor units, a number of virtual datasets at multiple granularities across a number of data dimensions, wherein the number of virtual datasets are generated by performing a series of data transformations on a set of data records;
- identifying, by the number of processor units, a number of data records in the set of data records using a virtual dataset that corresponds to a combination of data dimensions from the number of data dimensions;
- determining, by the number of processor units, a set of granularities for each of the combination of data dimensions for comparing the number of data records that were identified;
- and
- determining, by the number of processor units, a machine learning model to classify the number of data records based on probabilistic relationships among the combination of data dimensions;
- and
- classifying, by the number of processor units, the number of data records using the machine learning model and the set of granularities that were determined for the combination of data dimensions.
A computer system comprising:
- a hardware processor;
- and
- a classifier, in communication with a number of processor units, wherein the classifier is configured to execute program code for:
- generating a number of virtual datasets at multiple granularities across a number of data dimensions, wherein the number of virtual datasets are generated by performing a series of data transformations on a set of data records;
- identifying data records in the set of data records using a virtual dataset that corresponds to a combination of data dimensions from the number of data dimensions;
- determining a set of granularities for each of the combination of data dimensions for comparing the number of data records that were identified;
- and
- determining a machine learning model to classify the number of data records based on probabilistic relationships among the combination of data dimensions;
- and
- classifying the number of data records using the machine learning model and the set of granularities that were determined for the combination of data dimensions.
A computer program product comprising:
- a computer readable storage media;
- and
- program code, stored on the computer readable storage media, for classifying data records, and when executed, instruct a computer system to perform a method of:
- generating a number of virtual datasets at multiple granularities across a number of data dimensions, wherein the number of virtual datasets are generated by performing a series of data transformations on a set of data records;
- identifying data records in the set of data records using a virtual dataset that corresponds to a combination of data dimensions from the number of data dimensions;
- determining a set of granularities for each of the number of data dimensions for comparing the number of data records that were identified;
- and
- determining a machine learning model to classify the number of data records based on probabilistic relationships among the combination of data dimensions;
- and
- classifying the number of data records according to the machine learning model and the set of granularities that were determined for the combination of data dimensions.
Dependencies are inferred from the claim text.