Missing identification results

Describe the bug I have a custom entity A which with the regex ‘AAA’, and another custom entity B with the regex ‘BBB’ … custom entity D with the regex ‘DDD’. so my code like:

analyzer.registry.add_recognizer(PatternRecognizer(supported_entity='A', patterns=[Pattern('A', 'AAA', score=0.4)]))
analyzer.registry.add_recognizer(PatternRecognizer(supported_entity='B', patterns=[Pattern('B', 'BBB', score=0.4)]))
analyzer.registry.add_recognizer(PatternRecognizer(supported_entity='C', patterns=[Pattern('C', 'CCC', score=0.4)]))
analyzer.registry.add_recognizer(PatternRecognizer(supported_entity='D', patterns=[Pattern('D', 'DDD', score=0.4)]))
analyzer.analyze(text, 'en', score_threshold=0.5)

my problem is : I’m sure the text contains something that can be detected, even I only add one PatternRecognizer it’s work. But when I add multiple instances of PatternRecognizer, I can’t find the contents of the added PatternRecognizer as in the result.

Suppose I replace PatternRecognizer with CustomRecognizer, even if the analyzer.registry contains multiple instances of CustomRecognizer.

CustomRecongizer like:

class RegexRecognizer(EntityRecognizer):
    expected_confidence_level = 0.4

    def __init__(self, regex, entity, **kwargs):
        kwargs['supported_entities'] = [entity]
        kwargs['supported_language'] = 'zh'
        super().__init__(**kwargs)
        self.regex = regex
        self.entity = entity

    def load(self) -> None:
        pass

    def analyze(self, text: str, entities, nlp_artifacts):
        try:
            regex = re.compile(self.regex)
        except Exception:
            return []
        else:
            return [
                RecognizerResult(entity_type=self.entity,
                                 start=re_result.regs[0][0],
                                 end=re_result.regs[0][1],
                                 score=self.expected_confidence_level,
                                 analysis_explanation=AnalysisExplanation(self.entity, self.expected_confidence_level))
                for re_result in regex.finditer(text)
            ]

My NLP model is stanza, and every PatternRecognizer with different context. with the same text, if only recognizer:

{'港澳通行证': [{'content': 'CA0000001', 'end': 1339, 'score': 0.75, 'start': 1330},
           {'content': 'C00007565', 'end': 1349, 'score': 0.75, 'start': 1340}]}

there are many recognizer :

'港澳通行证': [{'content': 'CA0000001', 'end': 1339, 'score': 0.4, 'start': 1330},
           {'content': 'C00007565', 'end': 1349, 'score': 0.4, 'start': 1340}],

[omri374]

Yes, that’s exactly it. The bug arises from the fact that all recognizers have the same name. Passing a name would definitely fix it, but we’re looking into changing the implementation so that each recognizer would be treated individually. Makin the name mandatory would not be backward compatible.

[omri374]

Hi @hummingbird1989, yes this is likely the same bug. Potentially just changing the name of the recognizer would solve this.