Neural network systems and methods for generating distributed representations of electronic transaction information

Patent metadata

Field	Value
Publication number	US-10496997-B2
Application number	US-201816010305-A
Country	US
Kind code	B2
Filing date	Jun 15, 2018
Priority date	Jun 23, 2016
Publication date	Dec 3, 2019
Grant date	Dec 3, 2019

Abstract

Computer-implemented methods and systems are provided for generating a distributed representation of electronic transaction data. Generation may include receiving electronic transaction data including first and second entity identifiers. Generation may also include generating an output distributed representation by iteratively updating a distributed representation using the electronic transaction data. The distributed representation may include rows corresponding to first entity identifiers and rows corresponding to second entity identifiers. An iterative update may include generating a training sample and an embedding vector using the components and the distributed representation; determining, by a neural network, a predicted category from the embedding vector; and updating the distributed representation using the predicted category and the training sample. The embodiments may also include outputting the output distributed representation to determine authorization of electronic transactions. Disclosed embodiments may receive an electronic transaction and determine whether to authorize it based on a distributed representation of electronic transaction data.

First claim

What is claimed is:

1. An authorization system comprising:
at least one processor; and
at least one non-transitory computer readable medium containing instructions that when executed by the at least one processor cause the authorization system to perform operations comprising:
receiving electronic transaction data and historical transaction data, the electronic transaction data including an entity identifier component and an amount component of an electronic transaction;
determining a location of the electronic transaction in a space of a distributed representation comprising a mapping of electronic transaction components in a high-order space comprising first rows corresponding to entities and second rows corresponding to amount ranges based on the electronic transaction data;
determining locations of the historical transaction data in the distributed representation space;
determining a location of a first point in the distributed representation space based on averaging the locations of the historical transaction data;
determining a first distance from the location of the electronic transaction to the first point in the distributed representation space; and
authorizing the electronic transaction based on the first distance and the first point.

2. The authorization system of claim 1 , wherein determining the location of the first point based on the locations of the historical transaction data includes clustering the locations of the historical transaction data to determine central points, the central points including the first point.

3. The authorization system of claim 1 , wherein the historical transaction data is limited to electronic transactions including components satisfying one or more criteria.

4. The authorization system of claim 1 , wherein the historical transaction data includes one or more locations in the distributed representation space representative of previous electronic transactions, the one or more locations including the first point, and one or more corresponding distances, the one or more corresponding distances including the first distance.

5. The authorization system of claim 4 , wherein locations in the distributed representation space of a predetermined percentage of previous electronic transactions associated with the first point are closer to the first point than the first distance.

6. The authorization system of claim 1 , wherein determining the location of the electronic transaction includes:
identifying rows of the distributed representation, the identified rows including one of the first rows corresponding to a value of the entity identifier component and one of the second rows corresponding to a value of the amount component; and
determining the location of the electronic transaction as a linear combination of the identified rows.

7. The authorization system of claim 1 , wherein the electronic transaction data further includes at least one of an electronic transaction time component, a day of week component, a day of month component, and a day of year component.

8. The authorization system of claim 1 , the operations further comprising determining a fraud score based on fraud criteria and the distributed representation space.

9. The authorization system of claim 8 , wherein the fraud criteria include at least one of:
a card not present indication,
a geographic location associated with the entity identifier component, and
a predetermined category corresponding to the entity identifier component.

10. An authorization method performed by one or more authorization servers, comprising:
receiving an electronic transaction associated with a first entity and historical transaction data for the first entity;
receiving a distributed representation comprising a mapping of electronic transaction components in a high-order space, including first rows corresponding to entities and second rows corresponding to amount ranges;
determining a location of the electronic transaction in a space of the distributed representation based on a second entity associated with the electronic transaction and an amount associated with the electronic transaction;
determining locations of the historical transaction data in the distributed representation space;
determining a location of a first point in the distributed representation space based on averaging the locations of the historical transaction data; and
authorizing the electronic transaction using a decision rule, the location of the electronic transaction in the distributed representation space, and the location of the first point.

11. The authorization method of claim 10 , the method further comprising receiving contextual information associated with the first entity, the contextual information distinct from the historical transaction data, and wherein the location of the electronic transaction in the distributed representation space is further based on the contextual information.

12. The authorization method of claim 11 , wherein the contextual information includes an address of the first entity.

13. The authorization method of claim 10 , wherein the distributed representation includes third rows corresponding to distance ranges, and wherein the location of the electronic transaction in the distributed representation space is further based on a location associated with the electronic transaction.

14. The authorization method of claim 10 , wherein the electronic transaction is received from a system associated with the second entity, and wherein the location of the electronic transaction in the distributed representation space depends on values of a row of the distributed representation corresponding to the second entity.

15. The authorization method of claim 10 , the method further including mapping the amount to an amount range, and wherein determining the location of the electronic transaction in the distributed representation space includes determining a row of the distributed representation corresponding to the amount range.

16. The authorization method of claim 10 , wherein the electronic transaction is associated with a geographic area and the distributed representation corresponds to the geographic area.

17. The authorization method of claim 10 , wherein the decision rule partitions the distributed representation space into acceptable and unacceptable electronic transactions.

18. An authorization system comprising
at least one processor; and
at least one non-transitory computer readable medium containing instructions that when executed by the at least one processor cause the authorization system to perform operations comprising:
receiving data for an electronic transaction including values for multiple components;
receiving historical transaction data including sets of values for the multiple components;
determining an electronic transaction component location in a space of a distributed representation comprising a mapping of electronic transaction components in a high-order space for each of the multiple components based on the electronic transaction data;
determining a typical location in the distributed representation space for each of the multiple components based on averaging the sets of values for the multiple components;
determining differences between the electronic transaction component locations and the typical locations for each of the multiple components; and
authorizing the electronic transaction when the differences satisfy one or more authorization criteria.

Assignees

• Capital One Services Llc

Inventors

• Shami Mohammad
• Nedanov Bogdan
• Anstett Conor
• Edwards Joshua

Classifications

• G06N3/048

  Activation functions · CPC title

• G06N3/042

  Knowledge-based neural networks; Logical representations of neural networks · CPC title

• G06N5/046

  Forward inferencing; Production systems · CPC title

• G06N5/022

  Knowledge engineering; Knowledge acquisition · CPC title

• G06N3/084

  Backpropagation, e.g. using gradient descent · CPC title