Note: If you want to use a Dedicated AI Engine make sure you have purchased the feature before you start the training process.
In addition to using the Rossum validation screen for the purpose of dedicated AI engine training, there are three general rules that you must follow throughout the annotation process:
- Keep the annotations consistent: consistency is the main pillar of a successful training process. If data is present in multiple locations in an invoice, always annotate them in the same location. If other users are involved in your annotation process, it is especially important that you make sure they are in sync for the entire process and annotate the documents in the same way.
- Annotate data occurrences every time: to teach the engine to recognize specific data, it is important to annotate that data every time it occurs. By doing this, the dedicated AI engine will learn what the data looks like and where to search for it. Sometimes you may not need to have data extracted from one particular document while needing it extracted from another. For the training process, it is important to annotate the data’s occurrence every time.
- Use Magic Grid: when annotating line items, always predominantly use Magic Grid to the level when it is possible. If you have documents in which Magic Grid cannot be placed over all of the line items data you need to be extracted, use Magic Grid to extract the data where possible on the document, and extract the rest by point-and-click approach to achieve the best extraction accuracy results.
Following these rules will help optimize the accuracy of your dedicated AI engine.