Systems and methods for detecting out-of-pattern transactions

A computer-based method for detecting possible fraud by detecting out-of-pattern payment transactions, said method executed on a server having an inverse recommender module for receiving and detecting possibly fraudulent payment transactions, the inverse recommender module including a memory device for storing data and a processor communicatively coupled to the memory device and to a payment card processing network, the network routing communications between a plurality of merchants and a plurality of issuers of payment cards to account holders who have accounts with one or more of the issuers, each merchant being communicatively coupled to the processor via a merchant computing device where account holder payment transactions are initiated, the method comprising steps, performed by the processor, of:

• receiving first transaction data corresponding to historical payment transactions from a predetermined time period between the plurality of account holders and the plurality of merchants, wherein the transaction data includes, for each of the historical payment transactions, an identifier of the account associated with the respective account holder, an identifier of the respective merchant, an amount of purchase, and a date of purchase;
• linking, in the memory device using the first transaction data, the account of each account holder with pairs of merchants with which the account holder transacted business during the predetermined time period, wherein each pair of merchants in the linked data is a merchant pair;
• generating, from the linked data, a data structure comprising a merchant correspondence matrix for the predetermined time period, the merchant correspondence matrix comprising the plurality of merchants and a plurality of counters indicating a number of historical payment transactions between merchant pairs of the plurality of merchants and the plurality of account holders, wherein each counter is associated with a merchant pair and is configured to increment when, during the predetermined time period, an account holder transacts with both merchants of the merchant pair;
• storing the merchant correspondence matrix in the data structure of the memory device linking the merchant pairs to each account holder of the plurality of account holders;
• generating a historic transaction vector for each of the plurality of account holders using the historical payment transaction data of the respective account holder, wherein the historic transaction vector comprises a number of elements corresponding to a number of rows in the merchant correspondence matrix and each element corresponds to one of the plurality of merchants identified in the matrix, wherein the value of each element is set based on whether the account holder visited the corresponding merchant during the predetermined time period;
• receiving, from the merchant computing device of a merchant of the plurality of merchants, second transaction data associated with a new requested payment transaction between an account holder of the plurality of account holders and the merchant of the plurality of merchants, wherein the merchant comprises a new merchant with which the account holder has not previously transacted business;
• and
• generating an inverse recommender score for the new requested payment transaction with the new merchant based on the account holder historic transaction vector, including:
• applying the historic transaction vector to the merchant correspondence matrix to generate an inverse recommender score vector, wherein applying the historic transaction vector to the merchant correspondence matrix comprises inputting an amount of activation into the plurality of merchants in the inverse recommender score vector by multiplying the merchant correspondence matrix and the historic transaction vector a selectable number of times using a matrix-vector multiplication function, such that the amount of activation input to each merchant of the plurality of merchants is based on the number of previous visits to each merchant by the account holder;
• determining, for each element of the account holder historic transaction vector, a change between the element value of the account holder historic transaction vector and the value of the corresponding element of the inverse recommender score vector, wherein the change in the value for the element corresponding to the new merchant being significantly larger than the change in the values of the elements corresponding to others of the merchants indicates that the new requested transaction is out-of-pattern for the account holder;
• and
• alerting the new merchant, via communication with the merchant computing device, if the transaction is out-of-pattern for the account holder, wherein the alert is configured to cause the new merchant to either decline the requested transaction or flag it for further investigation.

One or more non-transitory computer-readable storage media having computer-executable instructions embodied thereon for detecting possible fraud by detecting out-of-pattern payment transactions using a server having an inverse recommender module for receiving and detecting possibly fraudulent payment transactions, the inverse recommender module including a memory device for storing data and at least one processor communicatively coupled to the memory device and to a payment card processing network, the network routing communications between a plurality of merchants and a plurality of issuers of payment cards to account holders who have accounts with one or more of the issuers, each merchant being communicatively coupled to the processor via a merchant computing device where account holder payment transactions are initiated, wherein when executed by the at least one processor, the computer-executable instructions cause the at least one processor to:

• receive first transaction data corresponding to historical payment transactions from a predetermined time period between the plurality of account holders and the plurality of merchants, wherein the transaction data includes, for each of the historical payment transactions, an identifier of the account associated with the respective account holder, an identifier of the respective merchant, an amount of purchase, and a date of purchase;
• link, in the memory device using the first transaction data, the account of each account holder with pairs of merchants with which the account holder transacted business during the predetermined time period, wherein each pair of merchants in the linked data is a merchant pair;
• generate, from the linked data, a data structure comprising a merchant correspondence matrix for the predetermined time period, the merchant correspondence matrix comprising the plurality of merchants and a plurality of counters indicating a number of historical payment transactions between merchant pairs of the plurality of merchants and the plurality of account holders, wherein each counter is associated with a merchant pair and is configured to increment when, during the predetermined time period, an account holder transacts with both merchants of the merchant pair;
• store the merchant correspondence matrix in the data structure of the memory device linking the merchant pairs to each account holder of the plurality of account holders;
• generate a historic transaction vector for each of the plurality of account holders using the historical payment transaction data of the respective account holder, wherein the historic transaction vector comprises a number of elements corresponding to a number of rows in the merchant correspondence matrix and each element corresponds to one of the plurality of merchants identified in the matrix, wherein the value of each element is set based on whether the account holder visited the corresponding merchant during the predetermined time period;
• receive, from the merchant computing device of a merchant of the plurality of merchants, second transaction data associated with a new requested payment transaction between an account holder of the plurality of account holders and the merchant of the plurality of merchants, wherein the merchant comprises a new merchant with which the account holder has not previously transacted business;
• and
• generate an inverse recommender score for the new requested payment transaction with the new merchant based on the account holder historic transaction vector, including:
• applying the historic transaction vector to the merchant correspondence matrix to generate an inverse recommender score vector, wherein applying the historic transaction vector to the merchant correspondence matrix comprises inputting an amount of activation into the plurality of merchants in the inverse recommender score vector by multiplying the merchant correspondence matrix and the historic transaction vector a selectable number of times using a matrix-vector multiplication function, such that the amount of activation input to each merchant of the plurality of merchants is based on the number of previous visits to each merchant by the account holder;
• determining, for each element of the account holder historic transaction vector, a change between the element value of the account holder historic transaction vector and the value of the corresponding element of the inverse recommender score vector, wherein the change in the value for the element corresponding to the new merchant being significantly larger than the change in the values of the elements corresponding to others of the merchants indicates that the new requested transaction is out-of-pattern for the account holder;
• and
• alerting the new merchant, via communication with the merchant computing device, if the transaction is out-of-pattern for the account holder, wherein the alert is configured to cause the new merchant to either decline the requested transaction or flag it for further investigation.