US 12,229,717 B2
Technologies for using machine learning to manage product catalogs
James Harris Hurst, Birmingham, AL (US); and David Berk, Las Vegas, NV (US)
Assigned to Fetch Rewards, LLC, Chicago, IL (US)
Filed by FETCH REWARDS, LLC, Chicago, IL (US)
Filed on Apr. 17, 2023, as Appl. No. 18/135,428.
Prior Publication US 2024/0346434 A1, Oct. 17, 2024
Int. Cl. G06Q 10/087 (2023.01); G06N 20/00 (2019.01)
CPC G06Q 10/087 (2013.01) [G06N 20/00 (2019.01)] 20 Claims
OG exemplary drawing
 
1. A computer-implemented method of using machine learning to manage a product datastore, the computer-implemented method comprising:
training, by at least one processor, a machine learning-based entity resolution model using a labeled dataset of known matches and non-matches of product records;
analyzing, by the at least one processor using a set of import logic, an additional dataset received from a data source, including:
identifying a product that is indicated in the additional dataset,
determining that the product does not match one of a set of products that exists in the product datastore, and
in response to determining that the product does not match one of the set of products:
determining that the set of import logic has identified the product a threshold amount of times, and
in response to determining that the set of import logic has identified the product the threshold amount of times, classifying the product for inclusion in the product datastore;
analyzing, by the at least one processor using a set of machine learning models, the additional dataset to create or update a data record associated with the product;
analyzing, by the at least one processor using the machine learning-based entity resolution model that was trained, the data record associated with the product, including:
determining that the product has an existing data record in the product datastore, and
identifying, from the data record, a set of data that is additive to the existing data record;
updating, by the at least one processor, the existing data record in the product datastore according to the set of data that was identified; and
re-training, by the at least one processor, the machine learning-based entity resolution model using additional training data comprising a set of resolutions made via a manual override component, wherein the machine learning-based entity resolution model that was re-trained is used for subsequent analyses of data records.