US 12,591,813 B1Granted

Using machine learning models to cause operator actions

AssigneeCitibank, N.A.

InventorsRamnath Mahale, Promiti Dutta, Thomas Joseph Snyder, Murli D. Buluswar

Life of this patent

Estimating the term…

Loading patent…

Abstract

Methods and systems are described for utilizing machine learning models to generate a response. The system receives an operator input requesting actions to prompt a desired user interaction. The system selects user-specific machine learning models trained on interactions associated with each user of a plurality of users. Based on the operator input, the system determines a model subset and selects a candidate operator action to test. The system inputs the candidate action into each model of the subset to obtain a predicted user interaction. The system then assesses whether the candidate action causes the models to predict the desired interaction. If the candidate action is effective, the system executes the candidate action in relation to the user subset.

Claims20

Dependent claims

A system for utilizing a plurality of machine learning models to generate a response, comprising:

a storage device;
and
one or more processors in a network system communicatively coupled to the storage device, wherein the one or more processors execute instructions that are stored in the storage device to cause the system to:
receive an operator input for one or more actions that cause a user interaction to be performed;
select, based on the operator input, a plurality of user-specific machine learning models, wherein each of the plurality of user-specific machine learning models is trained based on user interactions received as real time data from a corresponding user of a plurality of users and corresponding actions associated with the user interactions;
select, based on the operator input, a model subset of the plurality of user-specific machine learning models, wherein the model subset corresponds to a user subset of the plurality of users;
select a candidate operator action for the user subset;
input, into each user-specific machine learning model of the model subset of the plurality of user-specific machine learning models, the candidate operator action to cause each user-specific machine learning model of the model subset to predict a corresponding user interaction, wherein inputting the candidate operator action causes each user-specific machine learning model of the model subset to simulate execution of the candidate operator action within each user-specific machine learning model in real-time without actual execution;
determine whether the candidate operator action causes at least a threshold number of models of the model subset to predict the user interaction;
based on determining that the candidate operator action causes at least the threshold number of models of the model subset to predict the user interaction, generate, based on the candidate operator action, one or more instructions for executing the candidate operator action in relation to the user subset;
and
execute the candidate operator action in relation to each user of the user subset.

9 more elements

A method for utilizing a plurality of machine learning models to generate a response, the method comprising:

receiving an operator input for one or more actions that cause a user interaction to be performed;
selecting, based on the operator input, a plurality of user-specific machine learning models, wherein each of the plurality of user-specific machine learning models is trained based on user interactions associated with a corresponding user of a plurality of users;
selecting, based on the operator input, a model subset of the plurality of user-specific machine learning models;
selecting a candidate operator action for a user subset of the plurality of users corresponding to the model subset;
inputting, into each user-specific machine learning model of the model subset of the plurality of user-specific machine learning models, the candidate operator action to cause each user-specific machine learning model of the model subset to predict a corresponding user interaction, wherein inputting the candidate operator action causes each user-specific machine learning model of the model subset to simulate execution of the candidate operator action within each user-specific machine learning model in real-time without actual execution;
determining whether the candidate operator action causes at least a threshold number of models of the model subset to predict the user interaction;
and
based on determining that the candidate operator action causes at least the threshold number of models of the model subset to predict the user interaction, executing the candidate operator action in relation to the user subset.

5 more elements

One or more non-transitory computer-readable media storing instructions thereon, wherein the instructions cause one or more processors to perform operations comprising:

receiving an operator input for one or more actions that cause a user interaction to be performed;
selecting, based on the operator input, a plurality of user-specific machine learning models, wherein each of the plurality of user-specific machine learning models is trained based on user interactions associated with a corresponding user of a plurality of users;
selecting, based on the operator input, a model subset of the plurality of user-specific machine learning models;
selecting a candidate operator action for a user subset of the plurality of users corresponding to the model subset;
inputting, into each user-specific machine learning model of the model subset of the plurality of user-specific machine learning models, the candidate operator action to cause each user-specific machine learning model of the model subset to predict a corresponding user interaction, wherein inputting the candidate operator action causes each user-specific machine learning model of the model subset to simulate execution of the candidate operator action within each user-specific machine learning model in real-time without actual execution;
determining whether the candidate operator action causes at least a threshold number of models of the model subset to predict the user interaction;
and
based on determining that the candidate operator action causes at least the threshold number of models of the model subset to predict the user interaction, executing the candidate operator action in relation to the user subset.

5 more elements

Dependencies are inferred from the claim text.