Granted Patent B2

US 12,488,231 · App. 17/011,872 · Granted Dec 2, 2025

Inventors: Suraj Jayakumar (San Jose, CA); Amit Kumar Bansal (Sunnyvale, CA); Chao Cheng (Milpitas, CA)

Assignee: PAYPAL, INC.

G06N3/08G06F16/2474G06N3/044G06Q30/0185G06Q40/12G06Q40/03

Abstract

Claims (69)

1 . A system, comprising:

a non-transitory memory; and

one or more hardware processors coupled with the non-transitory memory and configured to read instructions from the non-transitory memory to cause the system to perform operations comprising:

accessing data associated with a user, the data being associated with a value for previous transactions conducted by the user;

accessing a neural network trained for predicting likelihoods of events occurring based on a model training using training data for a plurality of features from previous aggregate values from the previous transactions;

determining, for a set of one or more features of the plurality of features, an aggregate value, based on the data, for each feature of the set at one or more time intervals;

structuring a data table comprising time series data having the aggregate value for each feature and the one or more time intervals, wherein the data table is usable at an input layer of the neural network, and wherein the structuring comprises:

adding the aggregate value for each feature to the data table for the one or more time intervals over which the data for each feature is available, wherein the adding to the data table causes the data table to be structured for the input layer based on the model training;

inputting, by a decision engine, the time series data including the aggregate value for each feature to the input layer of the neural network;

generating, at an output layer of the neural network based on the model training, a predictive value indicating a predicted likelihood of an event occurring in association with the user;

tagging the user with a characteristic based on the predictive value and a first threshold value associated with the predicted likelihood of the event, wherein the characteristic is indicative of a risk of the event occurring with the user based on predictive value generated by the neural network;

executing, automatically without user input, an action responsive to a request of the user for a use of a digital account in association with the event, wherein the action prevents the use of the digital account or enables the use of the digital account;

generating, using the neural network, a combination of features that includes the predictive value from the output layer;

inputting the combination of features to a separate neural network different from the neural network, wherein the separate neural network utilizes the predictive value in the combination of features in place of the time series data;

generating an output of the separate neural network based on the inputting the combination of features; and

updating the action based on the output of the separate neural network and a second threshold value associated with the predicted likelihood of the event.

2 . The system of claim 1 , the operations further comprising:

storing a tag indicating the predictive value is at or above the first threshold or the second threshold in an account record associated with the user.

3 . The system of claim 1 , the operations further comprising:

receiving the request from the user in association with the use of the digital account with a merchant, for a loan request, or for an amount of credit, wherein the action comprises accepting or rejecting the request.

4 . The system of claim 1 , wherein the neural network is a long short-term memory (LSTM) neural network.

5 . The system of claim 1 , wherein the predictive value indicates a likelihood of the user defaulting on an obligation.

6 . The system of claim 1 , wherein the predictive value indicates a likelihood of the user responding to a message.

7 . The system of claim 1 , wherein the predictive value indicates a likelihood of the user being associated with a fraudulent activity.

8 . The system of claim 1 , wherein the plurality of features includes at least one of a transaction amount or a payment amount.

9 . A method, comprising:

accessing one or more data items associated with a user, each data item of the one or more data items being further associated with a data item value;

determining a neural network trained for predicting likelihoods of events occurring based on a model training using training data for a plurality of features from previous aggregate values for past data item values;

determining, for a set of one or more features of the plurality of features, an aggregate value, based on the one or more data items, for each feature of the set at one or more time intervals;

structuring a data table comprising time series data having the aggregate value for each feature and the one or more time intervals, wherein the data table is usable at an input layer of the neural network, and wherein the structuring comprises:

adding the aggregate value for each feature to the data table for the one or more time intervals over which the data item for each feature is available, wherein the adding to the data table causes the data table to be structured for the input layer based on the model training;

inputting each aggregate value to the input layer of the neural network;

generating, at an output layer of the neural network based on the model training, a predictive value associated with a likelihood of an event occurring with the user; and

tagging the user with a characteristic based on the predictive value and a first threshold value associated with the likelihood of the event, wherein the characteristic is indicative of a risk of the event occurring with the user based on the predictive value generated by the neural network;

executing, automatically without user input, an action responsive to a request of the user for a use of a digital account in association with the event, wherein the action prevents the use of the digital account or enables the use of the digital account;

generating, using the neural network, a combination of features that includes the predictive value from the output layer;

inputting the combination of features to a separate neural network different from the neural network, wherein the separate neural network utilizes the predictive value in the combination of features in place of the time series data;

generating an output of the separate neural network based on the inputting the combination of features; and

updating the action based on the output of the separate neural network and a second threshold value associated with the likelihood of the event.

10 . The method of claim 9 , wherein the inputting of each aggregate value comprises:

populating a matrix, each column of the matrix representing a feature of the set of one or more features and each row of the matrix corresponding to a time interval of the one or more time intervals, wherein each element of the matrix is populated by the aggregate value corresponding to the feature represented by the row and the time interval represented by the column at which the element is located; and

inputting the matrix to the input layer of the neural network.

11 . The method of claim 9 , further comprising:

compiling, from a database, a plurality of records, each record corresponding to a user account tagged with a characteristic, wherein the characteristic is associated with the predictive value;

determining, for each record of the plurality of records, an aggregate value, based on one or more data items associated with the record, for each feature of the set of one or more features at one or more time intervals associated with the record; and

training the neural network using the aggregate values based on the one or more data items associated with each record of the plurality of records.

12 . The method of claim 9 , further comprising:

storing an indication indicating the predictive value or the output is at or above the first threshold or the second threshold in an account record associated with the user.

13 . The method of claim 9 , further comprising:

receiving the request from the user in association with the use of the digital account with a merchant, for a loan request, or for an amount of credit, wherein the action comprises accepting, rejecting, or modifying the based on the predictive value or the output, whether to accept, reject, or modify the user request.

14 . The method of claim 9 , wherein the aggregate value for a feature at a time interval is a sum of every data item value associated with a data item associated with the feature, for every data item for which a time associated with the data item is within the time interval.

15 . The method of claim 9 , wherein the aggregate value for a feature at a time interval is a number of data items associated with the feature, for every data item for which a time associated with the data item is within the time interval.

16 . The method of claim 9 , wherein the aggregate value for a feature at a time interval is an arithmetic mean of every data item value associated with a data item associated with the feature, for every data item for which a time associated with the data item is within the time interval.

17 . A non-transitory machine-readable medium having stored thereon machine-readable instructions executable to cause a machine to perform operations comprising:

accessing data associated with a user, the data being further associated with a data value;

accessing a neural network trained for predicting likelihoods of events occurring based on a model training using training data for a plurality of features from previous aggregate values from previous transactions;

determining, for a set of one or more features of the plurality of features, an aggregate value, based on the data, for each feature of the set over one or more time intervals;

structuring a data table comprising time series data having the aggregate value for each feature and the one or more time intervals, wherein the data table is usable at an input layer of the neural network, and wherein the structuring comprises:

adding the aggregate value for each feature to the data table for the one or more time intervals over which the data for each feature is available, wherein the adding to the data table causes the data table to be structured for the input layer based on the model training;

inputting the time series including each aggregate value for each feature to the input layer of the neural network, wherein the neural network is configured to produce a predictive value a indicating a predicted likelihood of an event occurring in association with the user;

generating, at an output layer of the neural network based on the model training, the predictive value;

tagging the user with a characteristic based on the predictive value and a first threshold value associated with the predicted likelihood of the event, wherein the characteristic is indicative of a risk of the event occurring with the user based on predictive value generated by the neural network;

executing, automatically without user input, an action responsive to a request of the user for a use of a digital account in association with the event, wherein the action prevents the use of the digital account or enables the use of the digital account; and

storing a tag indicating the predictive value is at or above the threshold in an account record associated with the user.

18 . The non-transitory machine-readable medium of claim 17 , the operations further comprising:

receiving the request from the user in association with the use of the digital account with a merchant, for a loan request, or for an amount of credit, wherein the action comprises accepting, rejecting, or modifying the request.

19 . The non-transitory machine-readable medium of claim 17 , wherein the plurality of features includes at least one of a number of transactions or a number of payments.

20 . The non-transitory machine-readable medium of claim 17 , the operations further comprising:

standardizing each aggregate value.

Assignments (1)

Continuity (1)

References Cited (14)