US 11,669,899 B2
Credit decisioning based on graph neural networks
Mohammad Reza Sarshogh, Bellevue, WA (US); Christopher Bruss, Washington, DC (US); and Keegan Hines, Washington, DC (US)
Assigned to Capital One Services, LLC, McLean, VA (US)
Filed by Capital One Services, LLC, McLean, VA (US)
Filed on Dec. 20, 2021, as Appl. No. 17/556,397.
Application 17/556,397 is a continuation of application No. 16/857,780, filed on Apr. 24, 2020, granted, now 11,238,531.
Prior Publication US 2022/0114661 A1, Apr. 14, 2022
Int. Cl. G06Q 40/03 (2023.01); G06Q 10/10 (2023.01); G06N 3/08 (2023.01); G06N 3/04 (2023.01)
CPC G06Q 40/03 (2023.01) [G06N 3/04 (2013.01); G06N 3/08 (2013.01); G06Q 10/10 (2013.01)] 20 Claims
OG exemplary drawing
 
1. A computer-implemented method, comprising:
accessing a network graph comprising a plurality of nodes including a first node and a second node, wherein the network graph is based on data associated with a plurality of time intervals;
receiving, by a graph neural network executing on a processor at each of the plurality of time intervals, a respective message from each of a first subset of the plurality of nodes connected to the first node, the respective message received from each of the first subset comprising a respective embedding vector reflecting a respective current state of a respective node of the first subset at a respective one of the plurality of time intervals;
receiving, by the graph neural network at each of the plurality of time intervals, a respective message from each of a second subset of the plurality of nodes connected to the second node, the respective message received from each of the second subset comprising a respective embedding vector reflecting a respective current state of a respective node of the second subset at the respective one of the plurality of time intervals;
updating, by the graph neural network in a backward pass of the graph neural network at each of the plurality of time intervals, a plurality of weights for the first node and a plurality of weights for the second node;
receiving a first request at a first time interval of the plurality of time intervals, the first request associated with the second node;
processing, by the graph neural network based on a plurality of attributes associated with the second node at the first time interval, the plurality of weights as updated for the first node at the first time interval, the plurality of weights as updated for the second node at the first time interval, and the network graph at the first time interval to compute a credit decision to approve or deny a credit request, a score reflecting credit worthiness of a customer, an optimal credit limit, or an increased credit limit relative to an initial credit limit; and
storing credit decision to approve or deny the credit request, the score reflecting the credit worthiness of the customer, the optimal credit limit, or the increased credit limit as being associated with the second node at the first time interval.