What method does Azure form recognizer uses to train its model?

Question

I have different type of invoices, and tried training (train with label) each type of invoice in different model and compose it. But I am not able to get the required accuracy. So what I want to know as a customer/developer is, on what basis does azure train its model, so that I can train it properly.
Does it train on template basis, location basis or key value basis? Which type of unsupervised machine learning algorithm does it use to train its model?
I know I am asking a lot, but if it is possible to answer all my questions, please do.

Answer 1

Hi, Form Recognizer uses ML algorithms and OCR to analyze your forms, extract text and data, map field relationships as key-value pairs, and return a structured JSON output. The underlying algorithm used is proprietary and not shared publicly at this time. For best results, please ensure your training data set follows the input requirements. Provide at least five filled-in forms of the same type/structure. This document provides details on how to build a training dataset for custom model (review the additional tips). Furthermore, improve your model accuracy by labeling additional forms and retraining to create a new model (consider adding the prediction file to the training set, label it, and train again).

---

--- *Kindly Accept Answer if the information helps. Thanks.*