US 12,457,182 B2
Merging data from various content domains to train a machine learning model to generate predictions for a specific content domain
Alison Chi, McLean, VA (US); Cosette Goldstein, McLean, VA (US); Anirudha Simha, McLean, VA (US); Remel Tucker, McLean, VA (US); Nimesh Bernard, McLean, VA (US); and Ricky Su, McLean, VA (US)
Assigned to Capital One Services, LLC, McLean, VA (US)
Filed by Capital One Services, LLC, McLean, VA (US)
Filed on Jun. 9, 2021, as Appl. No. 17/342,732.
Prior Publication US 2022/0400159 A1, Dec. 15, 2022
Int. Cl. H04L 51/02 (2022.01); G06N 3/04 (2023.01); G06N 3/08 (2023.01)
CPC H04L 51/02 (2013.01) [G06N 3/04 (2013.01); G06N 3/08 (2013.01)] 20 Claims
OG exemplary drawing
 
2. A method for improving prediction model accuracy by training a prediction model for a specific content domain based on aggregated training data from disparate content domains, the method comprising:
obtaining vector representations of requests and solutions of (i) a first group of requests and solutions related to a first content domain and (ii) a second group of requests and solutions related to a second content domain, the second content domain being different from the first content domain;
supplementing the requests and solutions of the first group with at least a subset of the requests and solutions of the second group based on similarity criteria between (i) a first set of vector representations of requests and a first solution to the requests in the first group and (ii) a second set of vector representations of requests and a second solution to the requests in the second group to generate an aggregated group of requests and solutions;
providing the aggregated group of requests and solutions as training data to a first prediction model to cause the first prediction model to generate matching vector representations for requests within the aggregated group of requests and solutions;
providing a user input, obtained from a client device associated with a user, to the first prediction model to obtain a prediction of a solution for the user input; and
generating for display, the solution to the user on the client device.