US 11,922,286 B1
Recurrent neural networks to predict user requests
Brandon Scott Kotara, New York, NY (US); Gunjan Chandraprakash Vijayvergia, San Antonio, TX (US); Jon Eric Weissenburger, Frisco, TX (US); and Ben Hawkins, Plano, TX (US)
Assigned to United Services Automobile Association (USAA), San Antonio, TX (US)
Filed by United Services Automobile Association (USAA), San Antonio, TX (US)
Filed on Apr. 7, 2023, as Appl. No. 18/297,420.
Application 18/297,420 is a continuation of application No. 17/475,590, filed on Sep. 15, 2021, granted, now 11,645,494.
Application 17/475,590 is a continuation of application No. 15/910,325, filed on Mar. 2, 2018, granted, now 11,144,813, issued on Oct. 12, 2021.
Int. Cl. G06N 3/04 (2023.01); G06N 3/045 (2023.01); G06N 3/08 (2023.01); G06N 3/084 (2023.01); G06N 20/00 (2019.01); G06N 20/10 (2019.01); G06N 20/20 (2019.01)
CPC G06N 3/04 (2013.01) [G06N 3/045 (2023.01); G06N 3/08 (2013.01); G06N 20/00 (2019.01); G06N 20/10 (2019.01); G06N 20/20 (2019.01); G06N 3/084 (2013.01)] 20 Claims
OG exemplary drawing
 
8. A system comprising:
one or more processors; and
one or more memories storing instructions that, when executed by the one or more processors, cause the system to perform a process of training an auto-suggest model of recurrent neural networks, the process comprising:
receiving an input from a user via a user device;
obtaining a combined vector based on concatenating of a first vector and a second vector associated with the input,
wherein the first vector indicates a predicted next character to follow the input based on a string of characters in the input, and
wherein the second vector indicates a predicted next word to follow the input based on one or more words in the input;
inputting, via the auto-suggest model, the combined vector into at least one mathematical function to generate an output,
wherein the output includes an index value that indicates a location of an output solution for the input in a predetermined array of solutions;
selecting the output solution located at the index value, wherein the output solution includes a prediction of information requested via the user device;
sending the output solution to the user device;
receiving a response from the user device based on the output solution,
wherein the response indicates the user either selected the output solution or did not select the output solution; and
training the auto-suggest model by adjusting location sequencing of output solutions to the input based on the response from the user.