US 12,461,760 B2
Machine learning techniques for assessing interfaces
Atharv Bhat, McLean, VA (US); and Zhiyu Lin, McLean, VA (US)
Assigned to Capital One Services, LLC, McLean, VA (US)
Filed by Capital One Services, LLC, McLean, VA (US)
Filed on Nov. 10, 2023, as Appl. No. 18/507,002.
Prior Publication US 2025/0156199 A1, May 15, 2025
Int. Cl. G06F 9/451 (2018.01); G06F 3/04847 (2022.01); G06F 40/40 (2020.01); G06N 20/00 (2019.01); G06V 20/50 (2022.01); G06V 20/62 (2022.01)
CPC G06F 9/451 (2018.02) [G06F 3/04847 (2013.01); G06F 40/40 (2020.01); G06N 20/00 (2019.01); G06V 20/50 (2022.01); G06V 20/62 (2022.01)] 20 Claims
OG exemplary drawing
 
1. A system for assessing interfaces using machine learning models, the system comprising:
one or more processors; and
a non-transitory, computer-readable medium comprising instructions that, when executed by the one or more processors, causes operations comprising:
receiving a request for evaluating a user interface based on a plurality of parameters, wherein the request comprises (1) a representation of the user interface and (2) one or more user-defined target values for the plurality of parameters;
identifying, within the representation, a plurality of elements of the user interface and one or more structural properties of each element, wherein each structural property controls presentation of a corresponding element;
generating, for each element, (1) a corresponding composite embedding representing corresponding one or more structural properties of each element and (2) a corresponding content embedding representing content of each element;
generating a graph representation comprising a plurality of nodes and a plurality of edges, wherein each node corresponds to an element of the plurality of elements, wherein each node is assigned a vector representation defined by the corresponding composite embedding and the corresponding content embedding for that element, and wherein each edge comprises one or more values representing a relative distance between two elements of the plurality of elements;
generating, based on the graph representation, a graph embedding using an embedding model trained to transform graph representations into embeddings;
inputting the graph embedding into a parameter analysis machine learning model to obtain a corresponding predicted value for each of the plurality of parameters, wherein the parameter analysis machine learning model is trained to predict values for parameters based on graph embeddings; and
responsive to determining that the corresponding predicted value for one or more of the plurality of parameters do not match the one or more user-defined target values, generating sequentially, using a generative model, modified representations of the user interface until each corresponding predicted value of each of the plurality of parameters matches the one or more user-defined target values.