Method, system, and computer program product for improving machine learning models

1 . A computer-implemented method, comprising:

for each payment transaction of a plurality of payment transactions, generating, with a machine learning model of a machine learning system, an authorization decision based on transaction data associated with each of the plurality of payment transactions;

transmitting, with the machine learning system, the authorization decisions to a payment network processing the plurality of payment transactions;

forming, with the machine learning system, a data set comprising a plurality of data records, the data set comprising the transaction data and the authorization decisions associated with the plurality of payment transactions processed over the payment network, each data record associated with a payment transaction of the plurality of payment transactions, each of the plurality of payment transactions associated with an entity comprising a merchant;

transmitting, with the machine learning system, the data set to an active learning system comprising a plurality of subsystems, each subsystem comprising a separate machine learning model;

inputting, with the active learning system, the data set to a first subsystem of the plurality of subsystems, the first subsystem comprising a pre-trained first machine learning model;

based on the data set, generating, with the first machine learning model, a plurality of first embedding vectors, each respective first embedding vector of the plurality of first embedding vectors associated with a respective payment transaction of the plurality of payment transactions;

inputting, with the active learning system, the plurality of first embedding vectors to a second subsystem of the plurality of subsystems, the second subsystem comprising a second machine learning model;

based on the plurality of first embedding vectors, generating, with the second machine learning model, a plurality of second embedding vectors by mapping each respective first embedding vector of the plurality of first embedding vectors to a respective second embedding vector in a user-specific embedding space associated with a user;

inputting, with the active learning system, the plurality of second embedding vectors to a third subsystem of the plurality of subsystems, the third subsystem comprising a third machine learning model;

extracting, with the third machine learning model, feature data associated with at least one feature based on the plurality of second embedding vectors;

inputting, with the active learning system, an output from a machine learning system and the feature data to a fourth subsystem of the plurality of subsystems, the fourth subsystem comprising a fourth machine learning model;

based on the output from the machine learning system and the feature data, generating, with the fourth machine learning model, a relevance score for each payment transaction of the plurality of payment transactions;

determining, with the active learning system, a subset of payment transactions of the plurality of payment transactions based on the relevance score for each payment transaction of the subset of payment transactions;

communicating, with the active learning system, a feedback request to a user device of the user based on the subset of payment transactions;

receiving, with the active learning system, feedback data from the user device for each payment transaction of the subset of payment transactions, the feedback data comprising first data representing whether the user agrees with the authorization decision generated by the machine learning system for the payment transaction and second data representing a user's level of expertise for the payment transaction;

training, with the active learning system, at least one of the following: the second machine learning model, the third machine learning model, the fourth machine learning model, or any combination thereof based on the second data;

training, with the machine learning system, the machine learning model of the machine learning system based on the first data to form updated machine learning model;

receiving, with the machine learning system and from the payment network, an authorization request associated a subsequent payment transaction;

generating, with the updated machine learning model of the machine learning system, an authorization decision for the subsequent payment transaction based on transaction data associated with the subsequent payment transaction.

2 . The method of claim 1 , wherein training the at least one of the following: the second machine learning model, the third machine learning model, the fourth machine learning model, or any combination thereof comprises modifying at least one parameter of at least one of the following: the second machine learning model, the third machine learning model, the fourth machine learning model, or any combination thereof based on the second data.

3 . The method of claim 1 , wherein the first data comprises a binary classification for each payment transaction of the subset of payment transactions.

4 . The method of claim 1 , wherein the data set comprises multivariate time-series data comprising the plurality of data records, each respective data record of the plurality of data records comprising data elements based on a time sequence for a plurality of variables.

5 . The method of claim 1 , wherein training the at least one of the following: the second machine learning model, the third machine learning model, the fourth machine learning model, or any combination thereof comprises:

determining, with at least one processor, a loss based on the second data and a loss function; and

training, with at least one processor, the at least one of the following: the second machine learning model, the third machine learning model, the fourth machine learning model, or any combination thereof based on the loss.

6 . The method of claim 5 , wherein the loss function comprises at least one of the following:

a first component to widen a gap between pairs of payment transactions of the subset of payment transactions having feedback data associated with a correct relative ranking between payment transactions of each pair of payment transactions based on the relevance scores thereof;

a second component configured to narrow a gap between pairs of payment transactions of the subset of payment transactions having feedback data associated with an incorrect relative ranking between payment transactions of each pair of payment transactions based on the relevance scores thereof;

a third component configured to narrow a gap between payment transactions of the subset of payment transactions having feedback data associated with an incorrect relative ranking between the payment transactions of the subset of payment transactions and other payment transactions of the plurality of payment transactions not included in the subset of payment transactions;

or any combination thereof.

7 . The method of claim 6 , wherein the loss function further comprises a first hyperparameter for the first component, a second hyperparameter for the second component, and a third hyperparameter for the third component.

8 . The method of claim 1 , wherein the subset of payment transactions comprises a predetermined number of payment transactions having one of a highest relevance score or a lowest relevance score.

9 . A system comprising at least one processor configured to:

for each payment transaction of a plurality of payment transactions, generate, with a machine learning model, an authorization decision based on transaction data associated with each of the plurality of payment transactions;

transmit the authorization decisions to a payment network processing the plurality of payment transactions;

form a data set comprising a plurality of data records, the data set comprising the transaction data and the authorization decisions associated with the plurality of payment transactions processed over the payment network, each data record associated with a payment transaction of the plurality of payment transactions, each of the plurality of payment transactions associated with an entity comprising a merchant;

transmit the data set to an active learning system comprising a plurality of subsystems, each subsystem comprising a separate machine learning model;

input the data set to a first subsystem of the plurality of subsystems, the first subsystem comprising a pre-trained first machine learning model;

based on the data set, generate, with the first machine learning model, a plurality of first embedding vectors, each respective first embedding vector of the plurality of first embedding vectors associated with a respective payment transaction of the plurality of payment transactions;

input the plurality of first embedding vectors to a second subsystem of the plurality of subsystems, the second subsystem comprising a second machine learning model;

based on the plurality of first embedding vectors, generate, with the second machine learning model, a plurality of second embedding vectors by mapping each respective first embedding vector of the plurality of first embedding vectors to a respective second embedding vector in a user-specific embedding space associated with a user;

input the plurality of second embedding vectors to a third subsystem of the plurality of subsystems, the third subsystem comprising a third machine learning model;

extract, with the third machine learning model, feature data associated with at least one feature based on the plurality of second embedding vectors;

input an output from a machine learning system and the feature data to a fourth subsystem of the plurality of subsystems, the fourth subsystem comprising a fourth machine learning model;

based on the output from the machine learning system and the feature data, generate, with the fourth machine learning model, a relevance score for each payment transaction of the plurality of payment transactions;

determine a subset of payment transactions of the plurality of payment transactions based on the relevance score for each payment transaction of the subset of payment transactions;

communicate a feedback request to a user device of the user based on the subset of payment transactions;

receive feedback data from the user device for each payment transaction of the subset of payment transactions, the feedback data comprising first data representing whether the user agrees with the authorization decision generated by the machine learning system for the payment transaction and second data representing a user's level of expertise for the payment transaction;

train at least one of the following: the second machine learning model, the third machine learning model, the fourth machine learning model, or any combination thereof based on the second data;

train the machine learning model of the machine learning system based on the first data to form updated machine learning model;

receive, from the payment network, an authorization request associated a subsequent payment transaction;

generate, with the updated machine learning model, an authorization decision for the subsequent payment transaction based on transaction data associated with the subsequent payment transaction.

10 . The system of claim 9 , wherein training the at least one of the following: the second machine learning model, the third machine learning model, the fourth machine learning model, or any combination thereof comprises modifying at least one parameter of at least one of the following: the second machine learning model, the third machine learning model, the fourth machine learning model, or any combination thereof based on the second data.

11 . The system of claim 9 , wherein the first data comprises a binary classification for each payment transaction of the subset of payment transactions.

12 . The system of claim 9 , wherein the data set comprises multivariate time-series data comprising the plurality of data records, each respective data record of the plurality of data records comprising data elements based on a time sequence for a plurality of variables.

13 . The system of claim 9 , wherein training the at least one of the following: the second machine learning model, the third machine learning model, the fourth machine learning model, or any combination thereof comprises the at least one processor configured to:

determine a loss based on the second data and a loss function; and

train the at least one of the following: the second machine learning model, the third machine learning model, the fourth machine learning model, or any combination thereof based on the loss.

14 . The system of claim 13 , wherein the loss function comprises at least one of the following:

a first component to widen a gap between pairs of payment transactions of the subset of payment transactions having feedback data associated with a correct relative ranking between payment transactions of each pair of payment transactions based on the relevance scores thereof;

a second component configured to narrow a gap between pairs of payment transactions of the subset of payment transactions having feedback data associated with an incorrect relative ranking between payment transactions of each pair of payment transactions based on the relevance scores thereof;

a third component configured to narrow a gap between payment transactions of the subset of payment transactions having feedback data associated with an incorrect relative ranking between the payment transactions of the subset of payment transactions and other payment transactions of the plurality of payment transactions not included in the subset of payment transactions;

or any combination thereof.

15 . The system of claim 14 , wherein the loss function further comprises a first hyperparameter for the first component, a second hyperparameter for the second component, and a third hyperparameter for the third component.

16 . A computer program product, comprising at least one non-transitory computer-readable medium including program instructions that, when executed by a machine learning system and an active learning system, cause the machine learning system and the active learning system to:

for each payment transaction of a plurality of payment transactions, generate, with a machine learning model of the machine learning system, an authorization decision based on transaction data associated with each of the plurality of payment transactions;

transmit, with the machine learning system, the authorization decisions to a payment network processing the plurality of payment transactions;

form, with the machine learning system, a data set comprising a plurality of data records, the data set comprising the transaction data and the authorization decisions associated with the plurality of payment transactions processed over the payment network, each data record associated with a payment transaction of the plurality of payment transactions, each of the plurality of payment transactions associated with an entity comprising a merchant;

transmit, with the machine learning system, the data set to the active learning system comprising a plurality of subsystems, each subsystem comprising a separate machine learning model;

input, with the active learning system, the data set to a first subsystem of the plurality of subsystems, the first subsystem comprising a pre-trained first machine learning model;

based on the data set, generate, with the first machine learning model, a plurality of first embedding vectors, each respective first embedding vector of the plurality of first embedding vectors associated with a respective payment transaction of the plurality of payment transactions;

input, with the active learning system, the plurality of first embedding vectors to a second subsystem of the plurality of subsystems, the second subsystem comprising a second machine learning model;

based on the plurality of first embedding vectors, generate, with the second machine learning model, a plurality of second embedding vectors by mapping each respective first embedding vector of the plurality of first embedding vectors to a respective second embedding vector in a user-specific embedding space associated with a user;

input, with the active learning system, the plurality of second embedding vectors to a third subsystem of the plurality of subsystems, the third subsystem comprising a third machine learning model;

extract, with the third machine learning model, feature data associated with at least one feature based on the plurality of second embedding vectors;

input, with the active learning system, an output from the machine learning system and the feature data to a fourth subsystem of the plurality of subsystems, the fourth subsystem comprising a fourth machine learning model;

based on the output from the machine learning system and the feature data, generate, with the fourth machine learning model, a relevance score for each payment transaction of the plurality of payment transactions;

determine, with the active learning system, a subset of payment transactions of the plurality of payment transactions based on the relevance score for each payment transaction of the subset of payment transactions;

communicate, with the active learning system, a feedback request to a user device of the user based on the subset of payment transactions;

receive, with the active learning system, feedback data from the user device for each payment transaction of the subset of payment transactions, the feedback data comprising first data representing whether the user agrees with the authorization decision generated by the machine learning system for the payment transaction and second data representing a user's level of expertise for the payment transaction;

train, with the active learning system, at least one of the following: the second machine learning model, the third machine learning model, the fourth machine learning model, or any combination thereof based on the second data;

train, with the machine learning system, the machine learning model of the machine learning system based on the first data to form updated machine learning model;

receive, with the machine learning system and from the payment network, an authorization request associated a subsequent payment transaction;

generate, with the updated machine learning model of the machine learning system, an authorization decision for the subsequent payment transaction based on transaction data associated with the subsequent payment transaction.

17 . The method of claim 1 , further comprising:

determining, with the active learning system, that the payment transaction is relevant to the user from a plurality of users based on the second embedding vector in the user-specific embedding space.