I created a Custom Document Splitter with 78 labels and test it using some of my sample documents. I encountered a lot of low confidence score specially for labels with 10 or more pages. I looked at the results in the Version Details/View Full evaluation and I see different labels for each categories. EG. document_metrics:K1_Form_1, page_metrics_first_page:K1_Form_1, page_metrics_interior_page:K1_Form_1, for label/categ K1_Fomr_1. Is these what DocAI is using for page splitting, 1st page, interior page and last page? If that's case, can I intentionally remove some of the interior pages on my training docs to get a higher confidence score on my model? Can I also do this in my classification model as well? Thanks.
Al Monteagudo