Machine learning comparison of receipts and invoices
A legal spend management solution is described herein, a solution that utilizes improved machine learning algorithms to match lines of a legal invoice to lines in a receipt from a set of receipts. The matching uses cosine similarity algorithms and Levenshtein distances to determine whether there is a match between the receipt and the invoice lines. The machine learning results are displayed using a novel set of icons that present the confidence score with a set of three squares below a document icon.
1. A special purpose computer implemented method of visualizing a machine learning match of a receipt with a line of a document, the method comprising:
tokenizing the line of the document into a document vector of document tokens;
lemmatizing each token in the document vector of the document tokens, and storing the lemmatized document tokens in the document vector of the document tokens;
creating a document term frequency inverse document frequency vector for the document vector of the document tokens;
looping through one or more receipts in a set of receipts, reviewing each receipt, tokenizing a plurality of lines of the receipt into a receipt vector of receipt tokens, wherein the receipt vector includes a location indicator of the location of the receipt token in the receipt;
lemmatizing each receipt token in the receipt vector of the receipt tokens and storing the lemmatized receipt tokens in the receipt vector of the receipt tokens;
creating a receipt term frequency inverse document frequency vector for the receipt vector of the receipt tokens;
comparing the document term frequency inverse document frequency vector to the receipt term frequency inverse document frequency vector using a similarity algorithm to calculate a confidence score and storing the confidence score for each receipt;
determining a matching receipt by selecting the receipt with a highest confidence score;
displaying an indication of the highest confidence score with a variable icon;
displaying the receipt associated with the highest confidence score.
2. The method of claim 1 wherein the document is a legal invoice.
3. The method of claim 1 wherein the document is an expense report.
4. The method of claim 1 further comprising determining a Levenshtein Distance between an amount in the receipt line and an amount in the document line.
5. The method of claim 1 further comprising determining a Levenshtein Distance between a date in the receipt line and a date in the document line.
6. The method of claim 1 further comprising highlighting a line in the receipt with the highest confidence score.
7. The method of claim 1 wherein the variable icon displays a different number of items depending upon a magnitude of the highest confidence score.
8. The method of claim 7 wherein the variable icon displays three solid squares under a document icon based on one range of the highest confidence score.
9. The method of claim 7 wherein the variable icon displays one solid square under the document icon based on a second range of the highest confidence score.
10. The method of claim 1 further comprising comparing the document term frequency inverse document frequency vector to a billing guideline term frequency inverse document frequency vector using the similarity algorithm to calculate a guideline confidence score and storing the guideline confidence score; displaying the indication of the guideline confidence score with the variable icon.