Store Receipts to structured information

As the title suggests, this process is a tool to transform receipts into a table sheet. The process is made for receipts that are already scanned and processed with an OCR Tool.

Input: txt. Files

Output: table with the following columns: Date, price, category, receipt index, buyer, product description  

Roughly speaking this process is divided in the following steps:

1.    .txt2exset: In this sub process a receipt.txt file is segmented. Every line represents one example. So, the example set has as many   examples as there are lines in the original file. For every example (i.e. line) 2 regex extract the price: “(-?([0-9]+([\,\.][0-9]{2})))” and   “(((-?([0-9]|[0-9]{2}|[0-9]{3})([\,\.\'\;\s][0-9]{2}))(\s[ab]|[ab]|[12]|\s[12])))”

2.    Categorization: Each product should be categorized. To keep it simple I choose to build a dictionary of rules that stem products to a   category. Type: categoryX:.*product.keywordX.*  . To do this, the process has to transform the values from the data set to documents, loop these   documents and put it back into a data set.

3.    Manual corrections: Of course this procedure does not create a usable table. Manual corrections are necessary. That’s why the process   stores the example set several times in a sheet for manual correction. This means that the process has three break points. First (“Art und   Preis”) to correct the extracted prices and categories if necessary; second (“datum und geschäft”) to correct the extracted date and the   extracted store. This was a bit tricky because you have to pay attention to the continuity of the data set during the manual editing.

4.    Output: Because of OCR errors there are still many wrong characters in the extracted prices. You can correct these errors an format it in   the table sheet with a formula like this: “=GLÄTTEN(WECHSELN(WECHSELN(WECHSELN(WECHSELN(WECHSELN(W2;"b";"");"a";"");";";",");".";",");"'";","))”   and “=WERT(SÄUBERN(X2))” . At the end you may cross check the result with a pivot table.