I’ve developed an ML skill in the AI Center, and the pipeline shows a score above 90%. However, after deploying the ML Skill within the ML Extractor activity, no fields are being extracted — not even a single one. What could be the potential issue, and how can I resolve it?"
Hii check these once
- Skill Deployment & Connection Issues
- Ensure the ML Skill is successfully deployed in AI Center and is in the “Available” state.
- In your ML Extractor activity, verify that the ML Skill is correctly selected and linked.
- Incorrect Document Type Mapping
- Check if the Document Type in the Data Extraction Scope matches the training data format. If the structure differs, the model may fail to extract fields.
- Insufficient Training Data or Labeling Issues
- Even with a high validation score, if the training dataset lacks sufficient variability, the model may struggle with real-world documents.
- Verify that the training data includes diverse samples and that fields were correctly labeled during the Data Labeling process.
- Field Mapping in ML Extractor
- In the ML Extractor Trainer, ensure that fields are properly defined and mapped.
- Check if the fields in Taxonomy Manager align with the fields expected by the ML Extractor.
- Confidence Threshold Too High
- If the confidence threshold in the ML Extractor activity is set too high, the model may reject all extractions. Try lowering the threshold and test again.
- Pipeline Model Version
- If you’ve retrained the model multiple times, ensure you’re using the correct version of the ML Skill that aligns with the latest pipeline run.
- Document Understanding Framework Compatibility
- Confirm that the document format (PDF, Image, etc.) is compatible with the ML Extractor. Sometimes, pre-processing (e.g., OCR tuning) may be needed for better extraction
I’ve developed a document understanding flow in UiPath Studio (Desktop), created a project in AI Center, and performed data labeling. Based on that, I created an ML package and built a pipeline. The confidence score of my pipeline is 75%. After creating an ML Skill from the pipeline, I added it to the data extraction scope (ML Extractor) in the validation station. However, when I checked the status of the fields in the validation station, none of the fields are being extracted.
first try to create an evaluation pipeline and then add the file you are trying an see if you are able to see any data
now in ml extractor did you link the fields with taxonomy fields created?
is the file being used different from what you trained?
cheers
Have you configured the taxonomy correctly? Also, ensure you open “configure extractors,” check the box, and select all the fields you need to extract. Please refer to the attached screenshot for guidance.
1 Like
pikorpa
prashant1603765
Anil_G
ben.smith
jrdev2