US 12,321,330 B1
Entity resolution using machine learning
Wyatt Wade Berlinic, Cambridge, MA (US)
Assigned to Peregrine Technologies, Inc., San Francisco, CA (US)
Filed by Peregrine Technologies, Inc., San Francisco, CA (US)
Filed on Jul. 30, 2024, as Appl. No. 18/789,322.
Int. Cl. G06F 16/00 (2019.01); G06F 16/23 (2019.01)
CPC G06F 16/2329 (2019.01) 20 Claims
OG exemplary drawing
 
1. A method, comprising:
receiving a plurality of entity records;
extracting certain fields of the plurality of entity records into a version of the plurality of entity records;
using a first machine learning model to embed the version of the plurality of entity records into numerical embeddings having multidimensional numerical values representing the version of the plurality of entity records;
using the numerical embeddings to determine candidate relationships between the plurality of entity records;
filtering the numerical embeddings based candidate relationships using one or more filtering rules to determine a refined candidate relationships set for computational efficiency;
analyzing using a second machine learning model the refined candidate relationships set to determine one or more groups of related entity records included in the plurality of entity records, wherein each of the one or more groups of related entity records identify a corresponding common underlying entity; and
for a specific entity record group included in the determined one or more groups of related entity records, creating a new merged entity record that merges elements of individual related entity records included in the specific entity record group.