Functional limitations

We recommend limiting the number of document types per batch class to fifty or fewer. If a larger number of document variants is expected, the best practice is to merge the rules for the variants into fewer document types.

Computer vision is used to identify Key-Value pairs and is language agnostic. The Semantik AI engine is trained primarily with invoice-type documents and mixed results may occur with other document types. Best results are obtained when Key-Value pairs are in close proximity to one another.

Review table extraction rules to make sure they meet your requirements and refine them as needed. A couple of items to watch for:

• Simple tables are supported. Complex tables (those that include, but are not limited to, attributes such as multi-row headers and nested tables) may yield unexpected results.

• Tables may be falsely detected or not at all detected because of the nature of machine-learning models.

• There are no options to exclude columns or to select specific tables when using Semantik AI Engine.

Semantik AI Engine identifies tables by searching for visual elements such as headers, rows or visible/invisible boundaries, and other table-like structures.

Document Design Accelerator does not support Table Extraction in Transact 2022.1.00.

Table extraction results from Document Design Accelerator and Universal Document Automation may not match. Document Design Accelerator aids in the creation of conventional table extraction rules. Universal Document Automation uses the extraction results from the Semantik AI Engine.