Use the same setup for fields on different queues
Please use the same logic for all fields that represent the same fields on different queues. The fields can be set up using the Rossum’s Extraction schema editor.
If the same fields need to be extracted, they have to have the same field IDs and the same document type (number, string, date, or enum) in all queues.
The field labels can be different for each queue, but fields IDs need to remain the same. That way all annotations are connected to the same logic within the AI Engine and the accuracy is improving for all queues equally.
Please note that even if your understanding of some field seems to be different from the business point of view, we might considered them as the same for the purpose of training. If you have different queues for each languages and you are extracting the same data, please change only field labels, never field IDs.
If you are not sure, please contact Rossum’s support. We can compare your documents and give you some advice.
Adding new fields to a queue
New fields (meaning, capturing data from the invoice that you have not previously extracted) can be added to the new queue. It does not have any impact on queues that you are already using.
New fields will be trained during the next training.
We recommend to use a prefix for the line items fields item_XXXX, so that it is clear that the field is part of the table.
Let Rossum check the Extraction schema before annotating
Please let Rossum review the new schema before you start the first set of annotations for your new queue. Contact your Customer Success Manager and inform them about all the queues you want to add to your training.
Please inform your Customer Success Manager about all fields that are filled by some logic which is build on your side and need to stay in the schema. Please note that
rir_field_names should remain empty for these fields.
Don’t start with your annotations before your Customer Success Manager confirms that your schema is ready. Annotations are connected to field IDs and it would be problematic to change field IDs after the annotations have already been done.
If you already have a DE, we can also apply it to your new queue. This will help you with annotations, since you can get some predictions for already trained fields, which can speed up the process of annotating your first documents in the new queue.
Do Not Change
rir_field_names on Your Own
rir_fields_names for generic fields are prefilled automatically. For customs fields, they are filled by Rossum after the first DE training.
Again, for the same fields (meaning, fields you use to extract the same values in different queues), the
rir_fields_names need to stay the same for all queues with the same Dedicated Engine.