When using the ‘Classify Document’ activity from the ‘UiPath.Ephesoft.Activities’ package, some documents (between 2-5%) are returning different results compared to all the other methods I am aware of. This includes
- The ‘test classification’ function on transact,
- The API (/rest/ocrClassifyExtract),
- The document review function.
Is it possible to explain how the document classification confidence score is calculated?
The documentation mentions it’s a Lucene score? This seems to suggest it’s linked to the Search Classification plugin but this is disabled in the batch class we are using and when manually tested, provides a significantly different result.
For Example: When the same document is run through the various pathways the results are:
- Expected: Unknown (confidence = 0%)
- Test Classification: Unknown (0%)
- API: Unknown (0%)
- Review: Unknown (0%)
- Test Classification (with search classification plugin enabled): Type A (24%)
- This activity: Type A (88%)