Hi UiPath Experts,
I want your help!
I am using Document Understanding to extract data from Invoices. I’m using machine learning extractor. But in the present validation Station, I’m facing the below error.
And it also not identifying the fields.
Hi,
This error you’re encountering Present Validation Station—“Please select evidence for the field”— usually indicates that the Machine Learning Extractor is unable to confidently extract data for certain fields (in this case, the Document Type and others like Invoice Number, Date, etc.).
Here is a suggestion to try to fix the issue:
1. Check Field Mapping:
Ensure that the fields in your template are properly mapped to the correct document sections. If the extractor can’t locate evidence, it may be because the data in the document doesn’t match the expected layout.
2. Adjust Confidence Levels:
The error about “extracted data with confidence below 100” suggests that the extracted data doesn’t meet the minimum confidence threshold set for validation.
- You can try lowering the confidence threshold if it’s set too high.
- Go to the “Extraction” filter and adjust the confidence range to accept lower values.
3. Training the Machine Learning Model:
If this issue persists across multiple documents, the Machine Learning model might not be sufficiently trained to handle the variation in document formats.
- You can improve the model by re-training it using more annotated samples from the same or similar document types.
4. Manual Validation:
During the Present Validation Station, UiPath expects you to manually provide evidence for fields that weren’t extracted with high confidence. You can manually select the correct values from the document and provide the missing evidence. This data can then be used to improve future extractions.
5. Document Template Consistency:
Ensure that the structure of the documents being processed matches the document template you’ve configured for extraction. Any significant variation in layout can result in missing or incorrectly extracted fields.
Also try:
- Add additional training samples if your document structure varies.
- Adjust the confidence threshold to reduce manual validation when appropriate.
- Consider modifying the extraction logic (such as regex or keyword-based extraction) if needed to improve results for specific fields like invoice numbers and dates.
This could help fix the issue of missing field evidence and low-confidence extraction results.
As per error looks like the value is typed in the field instead indicate it on the document
cheers
please check the extraction results
This topic was automatically closed 3 days after the last reply. New replies are no longer allowed.