Transaction reconciliation and deduplication
Provided are systems and methods for reconciling transactions from different financial accounts via machine learning. In one example, a method may include storing a data file comprising a plurality of electronic transactions and a plurality of transaction records corresponding to the plurality of electronic transactions in memory, inputting the data file into a first machine learning model and identifying respective transaction attributes of the plurality of transaction records based on the input data file, determining, via execution of a second machine learning model, that a first transaction and a second transaction in the data file correspond to a common transaction based on attributes of the first and second transactions which are identified from respective transactions records of the first and second transactions, and modifying the data file to indicate that the first and second transactions correspond to the common transaction and storing the modified data file in the memory.
1 . A method comprising:
storing a data file comprising a plurality of transaction records corresponding to a plurality of electronic transactions in memory;
inputting the data file into a first machine learning model;
identifying, via execution of the first machine learning model, respective transaction attributes of the plurality of transaction records based on the input data file;
determining, via execution of a second machine learning model receiving as inputs thereto the transaction attributes of the plurality of transaction records identified by the execution of the first machine learning model, that a first transaction record and a second transaction record in the data file correspond to a common electronic transaction based on attributes of the first and second transaction records identified from the respective first and second transaction records via execution of the first machine learning model and a confidence score indicative of a level of confidence the second machine learning model associates with the determination that the first transaction record and the second transaction record in the data file correspond to the common electronic transaction;
modifying the data file to indicate, within a data structure thereof, that the first and second transaction records correspond to the common transaction, the modifying further including combining unique attributes from the first transaction record and the second transaction record into the data file;
storing the modified data file in the memory;
determining whether the confidence score is below an established threshold;
outputting, in an event the confidence score is determined to be below the established threshold, the first and second transaction records and the confidence score associated with the determination that the first transaction record and the second transaction record correspond to a common electronic transaction;
receiving an indication of an approval or disapproval of the determination that the first transaction record and the second transaction record in the data file correspond to a common electronic transaction;
generating at least one of a trained first machine learning model and a trained second machine learning model by training at least one of the first machine learning model and the second machine learning model, respectively, with training data including the received indication that the first transaction record and the second transaction record in the data file correspond to a common electronic transaction, the trained first machine learning model having an improved ability to identify transaction attributes of transaction records via an execution of the trained first machine learning model and the trained second machine learning model having an improved ability to determine whether multiple transaction records correspond to a common electronic transaction via an execution of the trained second machine learning model; and
executing a third machine learning model having an input thereto of the modified data file to verify an income with respect to the plurality of transaction records included in the modified data file.
2 . The method of claim 1 , wherein the identifying comprises estimating one or more of a date attribute, an amount attribute, and a counterparty attribute of the first transaction record via the execution of the first machine learning model on a transaction string included in the first transaction record.
3 . The method of claim 2 , wherein the determining comprises determining that the first transaction record and the second transaction are from the common transaction based on the estimated one or more of the date attribute, the amount attribute, and the counterparty attribute of the first transaction record and one or more of a date attribute, an amount attribute, and a counterparty attribute expressly included in the second transaction record.
4 . The method of claim 1 , wherein the identifying comprises estimating a counterparty entity attribute of the first transaction record via the execution of the first machine learning model on content within the first transaction record.
5 . The method of claim 4 , wherein the determining comprises determining that the first transaction record and the second transaction record are from opposing sides of the common transaction based on the counterparty identity attribute identified from the first transaction record via the execution of the first machine learning model and a payment source attribute expressly included in the second transaction record.
6 . The method of claim 1 , wherein the determining comprises determining that the first transaction record and the second transaction record are from the common transaction based on differing date attributes included in the first and second transaction records, respectively, via the execution of the second machine learning model.
7 . The method of claim 1 , wherein the determining comprises determining that the first transaction record and the second transaction record are from the common transaction based on differing payment amount attributes included in the first and second transaction records, respectively, via the execution of the second machine learning model.
8 . The method of claim 1 , wherein the method further comprises converting text from the plurality of transaction records into one or more vectors and inputting the one or more vectors into the first machine learning model during the execution of the first machine learning model.
9 . The method of claim 1 , wherein the determining comprises determining that the first and second transaction records are duplicate transaction records from two different sources, and the modifying comprises deleting one of the duplicate transaction records from the data file to create the modified data file.
10 . The method of claim 1 , wherein the determining comprises determining that the first and second transaction records comprise balancing credits and debits, and the modifying comprises aggregating attributes from the first and second transaction records into a single transaction record in the modified data file.
11 . A computing system comprising:
a memory configured to store a data file comprising a plurality of transaction records corresponding to a plurality of electronic transactions; and
a processor configured to:
input the data file into a first machine learning model:
identify, via execution of the first machine learning model, respective transaction attributes of the plurality of transaction records based on the input data file;
determine, via execution of a second machine learning model receiving as inputs thereto the transaction attributes of the plurality of transaction records identified by the execution of the first machine learning model, that a first transaction record and a second transaction record in the data file correspond to a common transaction based on attributes of the first and second transaction records identified from the first and second transaction records via the execution of the first machine learning model and a confidence score indicative of a level of confidence the second machine learning model associates with the determination that the first transaction record and the second transaction record in the data file correspond to the common electronic transaction,
modify the data file to indicate, within a data structure thereof, that the first and second transaction records correspond to the common transaction, the modifying further including combining unique attributes from the first transaction record and the second transaction record into the data file;
store the modified data file in memory;
determine whether the confidence score is below an established threshold;
output, in an event the confidence score is determined to be below the established threshold, the first and second transaction records and the confidence score associated with the determination that the first transaction record and the second transaction record correspond to a common electronic transaction;
receive an indication of an approval or disapproval of the determination that the first transaction record and the second transaction record in the data file correspond to a common electronic transaction;
generate at least one of a trained first machine learning model and a trained second machine learning model by training at least one of the first machine learning model and the second machine learning model, respectively, with training data including the received indication that the first transaction record and the second transaction record in the data file correspond to a common electronic transaction, the trained first machine learning model having an improved ability to identify transaction attributes of transaction records via an execution of the trained first machine learning model and the trained second machine learning model having an improved ability to determine whether multiple transaction records correspond to a common electronic transaction via an execution of the trained second machine learning model; and
execute a third machine learning model having an input thereto of the modified data file to verify an income with respect to the plurality of transaction records included in the modified data file.
12 . The computing system of claim 11 , wherein the processor is configured to estimate one or more of a date attribute, an amount attribute, and a counterparty attribute of the first transaction record via the execution of the first machine learning model on the first transaction record.
13 . The computing system of claim 12 , wherein the processor is configured to determine that the first transaction record and the second transaction record are from the common transaction based on the estimated one or more of the date attribute, the amount attribute, and the counterparty attribute of the transaction record of the first transaction and one or more of a date attribute, an amount attribute, and a counterparty attribute included in the second transaction record.
14 . The computing system of claim 11 , wherein the processor is configured to estimate a counterparty entity attribute of the first transaction record via the execution of the first machine learning model on content within the first transaction record.
15 . The computing system of claim 14 , wherein the processor is configured to determine that the first transaction record and the second transaction record are from opposing sides of the common transaction based on the counterparty identity attribute identified from the first transaction record via the execution of the second machine learning model and a counterparty attribute included in the second transaction record.
16 . The computing system of claim 11 , wherein the processor is configured to determine that the first transaction record and the second transaction record are from the common transaction based on differing date attributes included in the first and second transaction records, respectively, via the execution of the second machine learning model.
17 . The computing system of claim 11 , wherein the processor is configured to determine that the first transaction record and the second transaction record are from the common transaction based on differing payment amount attributes included in the first and second transaction records, respectively, via the execution of the second machine learning model.
18 . The computing system of claim 11 , wherein the processor is further configured to convert text from the plurality of transaction records into one or more vectors and input the one or more vectors into the first machine learning model during the execution of the first machine learning model.
19 . A non-transitory computer-readable medium comprising instructions which when executed by a processor cause a computer to perform a method comprising:
storing a data file comprising a plurality of transaction records corresponding to a plurality of electronic transactions in memory;
inputting the data file into a first machine learning model;
identifying, via execution of the first machine learning model, respective transaction attributes of the plurality of transaction records based on the input data file;
determining, via execution of a second machine learning model receiving as inputs thereto the transaction attributes of the plurality of transaction records identified by the execution of the first machine learning model, that a first transaction record and a second transaction record included in the data file correspond to a common transaction based on identified attributes of the first and second transaction records, identified from the first and second transaction records via the execution of the first machine learning model and a confidence score indicative of a level of confidence the second machine learning model associates with the determination that the first transaction record and the second transaction record in the data file correspond to the common electronic transaction;
modifying the data file to indicate that the first and second transaction records correspond to the common transaction, the modifying further including combining unique attributes from the first transaction record and the second transaction record into the data file;
storing the modified data file in the memory;
determining whether the confidence score is below an established threshold;
outputting, in an event the confidence score is determined to be below the established threshold, the first and second transaction records and the confidence score associated with the determination that the first transaction record and the second transaction record correspond to a common electronic transaction;
receiving an indication of an approval or disapproval of the determination that the first transaction record and the second transaction record in the data file correspond to a common electronic transaction;
generating at least one of a trained first machine learning model and a trained second machine learning model by training at least one of the first machine learning model and the second machine learning model, respectively, with training data including the received indication that the first transaction record and the second transaction record in the data file correspond to a common electronic transaction, the trained first machine learning model having an improved ability to identify transaction attributes of transaction records via an execution of the trained first machine learning model and the trained second machine learning model having an improved ability to determine whether multiple transaction records correspond to a common electronic transaction via an execution of the trained second machine learning model; and
executing a third machine learning model having an input thereto of the modified data file to verify an income with respect to the plurality of transaction records included in the modified data file.