How model training works
Veryfi's Machine Learning models undergo frequent training.
Veryfi includes all changes users make to processed documents into constant training sessions.
How to help Veryfi improve the accuracy of data extraction?
All you need to do is to correct any issues you spot through the PUT, POST, and DELETE operations. Make sure documents you update with PUT remain inside your Veryfi account at least the next 2-3 weeks after you updated it. Veryfi releases a new and improved version of the model every 1-2 weeks.
Here is what users can update:
Update an existing document (vendor, total, tax, bill_to, etc.)
Update existing line item on a document (total, tax, price, category, qty, etc.)
Add a line item to an existing document
Delete a line item from the existing document
Please find detailed instructions and query parameters samples inside your Veryfi API Web Hub.
Still get same errors?
If you still face difficulties with a new model not reading particular parts of the documents, we will be happy to review your case.
Please, contact support@veryfi.com. Provide details, context, attach the original file and JSON response. You must highlight the exact parameter Veryfi fails to extract. Veryfi tech team will be happy to review your specific case.
Accuracy depends on document type, image quality, the invoice layout, structure, language, vendor, etc. We ask you to send the original file, so we could check how good Veryfi is at reading this particular type of document. In case if we don't have enough similar data, we will ask you for more samples to annotate and use for model training. Having 20+ similar documents will help. Send us 20-100 documents, and we will be happy to submit them for manual annotation and training. Thus, working from both sides, we achieved accuracy improvement faster.
Need help?
If you run into any issue or need help, please contact support@veryfi.com
Other Related Articles:
Please, reach out to support@veryfi.com if you have any questions.