Goal

Develop a solution based on neural networks, which will automatically extract the necessary data from payment orders, such as the name of the paid service, the name of the individual, the payment date.

Our solution

We built neural network model of the Russian language by training a neural network based on the language corpus.

To solve this particular task, we conducted additional neural network training on the actual payment documents received from the Customer.

We marked up more than 20 000 real payment documents using regular expressions to create a model for finding the information in the document.

As the result, we obtained a neural network able to quickly and accurately extract the needed data from the documents.

Results

More than 98% of documents are processed automatically.

Documents containing no spaces, spelling errors, and scan artifacts are also handled successfully.