Back to results
US 12,182,830 B2Granted
Method, system, and non-transitory computer-readable recording media for supporting payment
AssigneeBank X Inc.
InventorSeung Hon Kim
Life of this patent
Estimating the term…
Abstract
A method for supporting payment is provided. The method includes the steps of: acquiring information on at least one payment means associated with a user and payment information of the at least one payment means; and determining a payment means to recommend to the user at the time of payment by referring to the result of learning performed using learning data including feature information extracted from the information on the at least one payment means and the payment information of the at least one payment means to maximize cumulative benefits according to the user's payment pattern.
Claims17
A method for supporting payment, comprising:
- acquiring information on at least one payment means associated with a user and payment information of the at least one payment means;
- determining, using an artificial neural network based learning model and using reinforcement learning, a policy that maximizes a value of an action corresponding to a use of the at least one payment means by the user based on a state corresponding to feature information;
- training the learning model using learning data including the feature information extracted from the information on the at least one payment means and the payment information of the at least one payment means by maximizing cumulative benefits according to a payment pattern of the user and by maximizing the value of the action corresponding to the use of the at least one payment means by the user based on to the state corresponding to the feature information, the value of the action being determined based on an expected value of a sum of a current reward and a future reward corresponding to the cumulative benefits according to the payment pattern of the user,
- wherein the reinforcement learning is based on Q-learning for determining the policy that maximizes the value;
- updating, using the Q-learning, a value of Q for a specific state-action pair by repeatedly adding a reward received for the specific state-action pair and (a maximum Q value)×(a discount rate) in an adjacent state, to determine the policy that maximizes the value of the action, the discount rate indicating a size of a benefit in a long-term perspective;
- determining a payment means to recommend to the user at a time of payment based on the determined policy that maximizes the value of the action, the trained learning model, and the updated value of Q;
- and
- providing notification of the recommended payment means to the user.
5 more elements
A system for supporting payment, comprising:
- a memory configured to store computer program;
- and
- a processor configured to execute the computer program to perform:
- acquiring information about at least one payment means associated with a user and payment information of the at least one payment means;
- determining, using an artificial neural network based learning model and using reinforcement learning, a policy that maximizes a value of an action corresponding to a use of the at least one payment means by the user based on a state corresponding to feature information;
- training the learning model using learning data including the feature information extracted from the information about the at least one payment means and the payment information to determine a payment means by maximizing that maximizes cumulative benefits according to a payment pattern of the user and by maximizing the value of the action corresponding to the use of the at least one payment means by the user based on the state corresponding to the feature information, the value of the action being determined based on an expected value of a sum of a current reward and a future reward corresponding to the cumulative benefits according to the payment pattern of the user,
- wherein the reinforcement learning is based on Q-learning for determining the policy that maximizes the value;
- updating, using the Q-learning, a value of Q for a specific state-action pair by repeatedly adding a reward received for the specific state-action pair and (a maximum Q value)×(a discount rate) in an adjacent state, to determine the policy that maximizes the value of the action, the discount rate indicating a size of a benefit in a long-term perspective;
- determining a payment means to recommend to the user at a time of payment based on the determined policy that maximizes the value of the action, the trained learning model, and the updated value of Q;
- and
- providing notification of the recommended payment means to the user.
8 more elements
Dependencies are inferred from the claim text.