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

Patent metadata

Field	Value
Publication number	US-2017372318-A1
Application number	US-201715632158-A
Country	US
Kind code	A1
Filing date	Jun 23, 2017
Priority date	Jun 23, 2016
Publication date	Dec 28, 2017
Grant date	—

Abstract

Computer-implemented methods and systems are provided for generating a distributed representation of electronic transaction data. Consistent with disclosed embodiments, 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 also receive an electronic transaction and determine whether to authorize the electronic transaction based on a distributed representation of electronic transaction data.

First claim

What is claimed is:

1 . A computer-implemented neural network method for generating a distributed representation of electronic transaction data, comprising:
receiving electronic transaction data, the electronic transaction data including components of an electronic transaction, the components including a first entity identifier and a second entity identifier; generating, by at least one processor, an output distributed representation by iteratively updating a distributed representation using the electronic transaction data, the distributed representation including rows corresponding to a plurality of first entity identifiers and rows corresponding to a plurality of second entity identifiers, an iteration comprising:
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; and
outputting the output distributed representation to determine authorization of electronic transactions.

2 . The computer-implemented method of claim 1 , wherein one of the components is selected as the training sample.

3 . The computer-implemented method of claim 1 , wherein a length of the embedding vector equals a length of the rows of the distributed representation.

4 . The computer-implemented method of claim 1 , wherein generating the embedding vector comprises averaging rows of the distributed representation corresponding to the components.

5 . The computer-implemented method of claim 1 , wherein updating the distributed representation comprises:
calculating an update vector representing the effect on a cost function of changing the embedding vector, and updating the rows of the distributed representation corresponding to the components using the update vector.

6 . The computer-implemented method of claim 5 , wherein a length of the update vector equals a length of the rows of the distributed representation.

7 . The computer-implemented method of claim 5 , wherein the update vector contributes equally to the rows of the distributed representation corresponding to the components.

8 . The computer-implemented method of claim 1 , wherein the components further include a continuous variable, and generating the training sample and the at least one embedding vector comprises discretizing the continuous variable.

9 . The computer-implemented method of claim 7 , wherein the continuous variable is an electronic transaction time or an electronic transaction amount, and wherein the components further include at least one of electronic transaction day-of-week, day-of-month, and day-of-year.

10 . The computer-implemented method of claim 1 , further comprising determining first entity identifier frequencies in the electronic transaction data, and wherein generating the final distributed representation further includes resampling the electronic transaction data based on the first entity identifier frequencies.

11 . The computer-implemented method of claim 1 , wherein the distributed representation includes between 80 and 120 features, and the neural network comprises a hidden layer including between 80 and 120 nodes.

12 . The computer-implemented method of claim 1 , wherein generating the output distributed representation further comprises:
identifying common rows in the distributed representation and another distributed representation; aligning the distributed representation and the other distributed representation using values of the common rows; and combining the distributed representation and the other distributed representation.

13 . The computer-implemented method of claim 12 , wherein aligning the distributed representation and the other distributed representation comprises determining a transformation that minimizes a function of the values of the common rows, and wherein the distributed representation and other distributed representation are combined using the transformation.

14 . The computer-implemented method of claim 12 , wherein the other distributed representation was generated using other electronic transaction data, and wherein the electronic transaction data and the other electronic transaction data correspond to at least one of differing time periods, geographic areas, and second entity demographic groups.

15 . The computer-implemented method of claim 12 , wherein the common rows correspond to first entity identifiers.

16 . An authorization server 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 server to perform operations comprising:
receiving components of an electronic transaction from a first system, the components including a first entity identifier and a second entity identifier,
generating a representation of the electronic transaction using the components and a distributed representation of electronic transaction data including rows corresponding to the components,
determining authorization of the electronic transaction by applying a decision rule that uses:
the representation of the electronic transaction, and
representations of past electronic transactions associated with at least one of the merchant identifier and the second entity identifier; and

providing an authorization indication to the first system based on the determined authorization.

17 . The authorization server of claim 16 , wherein the first entity identifier corresponds to a first entity associated with the first system.

18 . The authorization server of claim 16 , wherein applying the decision rule comprises determining whether a distance between the representation of the electronic transaction and a point dependent on the representations of past electronic transactions exceeds a value.

19 . The authorization server of claim 16 , wherein generating the representation of the electronic transaction comprises:
determining a first row of the distributed representation of electronic transaction data corresponding to the first entity identifier; determining a second row of the distributed representation of electronic transaction data corresponding to the second entity identifier; generating the representation of the electronic transaction as an average of at least the first row and the second row.

20 . The authorization server of claim 16 , wherein the components further include a continuous variable, and wherein generating the representation of the electronic transaction further comprises discretizing the continuous variable.

21 . The authorization server of claim 16 , wherein the components further include at least one of time of electronic transaction, amount of electronic transaction, day of week, day of month, and day of year.

22 . The authorization server of claim 16 , wherein the decision rule additionally uses fraud criteria, the fraud criteria including at least one of:
a card not present indication, a first geographic location associated with the first entity identifier, a second geographic location associated with the second entity identifier, and a predetermined category corresponding to the first entity identifier.

23 . An authorization server comprising:
at least one processor; and at least one non-transitory computer readable medium containing

Assignees

• Capital One Services Llc

Inventors

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

Classifications

• G06N3/042

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

• G06N3/048

  Activation functions · CPC title

• G06N3/084

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

• G06N5/046

  Forward inferencing; Production systems · CPC title

• G06N5/022

  Knowledge engineering; Knowledge acquisition · CPC title