US 12,260,462 B2
Real-time expense auditing and machine learning system
Winston Leonard Wang, San Francisco, CA (US); Parker Ralph Kuncl, Seattle, WA (US); Kelly Bailey, San Francisco, CA (US); and Matthew Brigante, Seattle, WA (US)
Assigned to Oracle International Corporation, Redwood Shores, CA (US)
Filed by Oracle International Corporation, Redwood Shores, CA (US)
Filed on Sep. 10, 2020, as Appl. No. 17/017,380.
Claims priority of provisional application 62/898,699, filed on Sep. 11, 2019.
Prior Publication US 2021/0073920 A1, Mar. 11, 2021
Int. Cl. G06Q 40/12 (2023.01); G06F 40/30 (2020.01); G06N 5/04 (2023.01); G06N 20/00 (2019.01); G06Q 10/105 (2023.01)
CPC G06Q 40/12 (2013.12) [G06N 5/04 (2013.01); G06N 20/00 (2019.01); G06Q 10/105 (2013.01); G06F 40/30 (2020.01)] 23 Claims
OG exemplary drawing
 
1. One or more non-transitory machine-readable media storing instructions which, when executed by one or more processors, cause:
training, by an expense auditing system, a neural network to compute audit risk scores as a function of expense descriptions, wherein training the neural network comprises performing adjustments to memory parameters of one or more cells in the neural network that include a memory to learn temporal dependencies based on an order in which feature vectors representing a sequence of expense activity are fed through the neural network, wherein the memory of a cell serves as a feedback connection that causes a first output of the cell to affect a contribution of a second output of the cell to the audit risk scores;
receiving, by the expense auditing system through an application programming interface from one or more external data sources, a set of expenses associated with an employee;
after training the neural network, receiving, by the expense auditing system through an intelligent agent interface, a query about an expense that includes an expense description associated with the expense;
responsive to receiving the query through the intelligent agent interface, identifying, by the expense auditing system within the set of expenses received from the one or more external data sources, at least one other expense incurred within a threshold timeframe from the expense and generating a set of feature vectors including a first feature vector representing the expense that includes values based on attributes extracted from the expense description and at least one other feature vector representing the at least one other expense incurred within the threshold timeframe from the expense;
computing, by the expense auditing system using the trained neural network, a first audit risk score associated with the expense description, wherein the trained neural network estimates the first audit risk score by performing forward propagation, using a sequence of feature vectors including the first feature vector representing the expense and the at least one other feature vector representing the at least one other expense in an order in which the expense and the at least one other expense occurred, based at least in part on the memory parameters associated with the one or more cells in the neural network and at least one learned temporal dependency between the expense description and a previous expense activity associated with the employee;
comparing, by the expense auditing system, the first audit risk score with an audit trigger comprising one or more conditions that, when satisfied, identifies expense descriptions that are at risk of being audited;
determining, by the expense auditing system, that the first audit risk score satisfies the audit trigger; and
responsive to determining that the first audit risk score satisfies the audit trigger: alerting, by the expense auditing system, the employee that the expense description is at risk of being audited.