Specific versus generic learning

The concept of specific versus generic learning only applies to Extraction.

In specific learning, TotalAgility first classifies a document and then applies the learned position and keywords for that specific internal document type to extract the data. Specific learning is typically used for invoices and documents alike. While there is only one actual "Invoices" document type in the project, internally the trainable locators (see next section) classify the individual vendor layouts. Once they know the layout, they can locate the data. Specific training is precise, but the learned knowledge cannot be applied to unknown documents.

Generic learning is independent of the document layout. TotalAgility learns the keywords and other properties associated with the values on the sample documents and then applies this knowledge to documents and layouts it has not seen before. This is less precise than specific learning, but more generic, as the name indicates.

Some locators support both generic and specific extraction and usually fallback. They first try to find the data using the specific knowledge, but if they do not recognize that specific layout, they try to apply the generic knowledge. See the next topic.