Same Analyzer detects entity in text but not in image

[NuiMrme]

Describe the bug same Analyzer detects LOCATION entity token in text but fails to detect the same token in an image

To Reproduce

analyzer=AnalyzerEngine(nlp_engine=nlp_engine_with_french,
    log_decision_process="true",supported_languages = ["fr","en"])

print(analyzer.analyze(text='VALENCE', language ="en"))

ImageAnalyzer = ImageAnalyzerEngine(analyzer_engine = analyzer)
engine = ImageRedactorEngine(image_analyzer_engine = ImageAnalyzer)

Expected behavior VALENCE is detected as location, even if I change the language, the text, lower-case etc... it is detected as LOCATION. If I use the same Analyzer to create an ImageAnalyzer, VALENCE should be detected as LOCATION if the word is there in the image.

[omri374]

Could it be the the OCR engine doesn't recognizer this text? Have you tried running tesseract on it to see the output?

[NuiMrme]

since I have the log_decision_process to "true", the word "VALENCE" is there in the log

Edit: That being said, I created an empty image with just the word "VALENCE" on it, and it was detected as LOCATION. Does the detection depends on the words before and after ??

[omri374]

Yes, location is detected using a named entity recognition model. context words could certainly change the output. If you have a finite list of locations, you can create a deny list and pass it to the analyzer engine.

[NuiMrme]

Yes, location is detected using a named entity recognition model. context words could certainly change the output. If you have a finite list of locations, you can create a deny list and pass it to the analyzer engine.

Thanks for your reply. I actually have that already in my code but I still don't detect this one location. I think the problem is deeper, something about that exact document makes it problematic. In my deny list though I have the locations first letter Capital, in the document it is written all in CAPITAL letters, not sure if this is a problem.