Methods and systems for automatically detecting fraud and compliance issues in expense reports and invoices

1. A method of detecting anomalies in expense reports, comprising:

applying a semantic analysis mechanism to an expense report submitted by an employee, wherein the semantic analysis mechanism includes instructions executed by a processor to:

build a set of data structures from expense report data,

provide the set of data structures from the expense report data to a probabilistic information engine that uses an expense ontology, wherein the expense ontology defines a set of categories of expenses and expense attributes, and

determine with the probabilistic information engine at least one transaction expense type in the expense report based on the expense topology to form a first enhancement of the expense report data;

supplementing the first enhancement of the expense report data with data obtained from a World Wide Web source to form a second enhancement of the expense report data; and

executing machine learning models on the second enhancement of the expense report data, wherein the machine learning models include:

a first trained machine learning model to detect employee interaction with a corrupt entity, the first trained machine learning model is trained with Politically Exposed Person (PEP) data, fictional person data and foreign location risk data,

a second trained machine learning model to detect employee activity indicative of bribery, the second trained machine learning model is trained with PEP data, fictional person data and foreign location risk data,

a third trained machine learning model to detect violation of company travel guidelines, the third trained machine learning model is trained with travel class policies and travel booking provider data,

a fourth trained machine learning model to detect an expense request inconsistent with historic expense data, the fourth trained machine learning model is trained using historical risk and behavioral data for employees, and

a fifth trained machine learning model to detect employee interaction with a vendor associated with fraudulent activities, the fifth trained machine learning model is trained using vendor profile data, fraudulent vendor data, vendor address data and fictional vendor data;

wherein one or more of the machine learning models utilize a Support Vector Machine (SVM) linear kernel model, wherein the SVM linear kernel model is trained using a specified number of training samples of employees in a company to be memory efficient.

2. The method of claim 1 , wherein the expense report data comprises expense report textual data including a user justification of expenses, merchant names, attendee names and locations.

3. The method of claim 2 , wherein the expense report data includes an expense cost in foreign currency, an expense cost in home currency, an exchange rate and an expense date.

4. The method of claim 1 , wherein at least one machine learning model uses a kernel algorithm.

5. The method of claim 1 , wherein at least one machine learning model uses a logistic regression algorithm.

6. The method of claim 1 , wherein at least one machine learning model uses a radial basis kernel algorithm.

7. The method of claim 1 , wherein at least one machine learning model uses K-nearest neighbor clustering.

8. The method of claim 1 , further comprising:

scanning a set of receipts associated with the expense report,

using optical character recognition to form scanned receipts, and

matching a transaction amount in the expense report with a scanned receipt.

9. The method of claim 1 further comprising instructions executed by the processor to identify a previously processed expense in the expense report.