Hi, I am having a serious issue with rasa nlu where entities are not identifying correctly.But the weird scenario is that it works(identify the entity) when i retrain the NLU model. In times when i edit the nlu_data.md file and retrain the model the entity wont work then i would have to retrain the model again and again the model to make it identify each entity.I find the model very inconsistent and least confident.
My NLU pipeline
language: “en” pipeline:
- name: “tokenizer_whitespace”
- name: “intent_entity_featurizer_regex”
- name: “ner_crf”
- name: “ner_synonyms”
- name: “intent_featurizer_count_vectors”
- name: “intent_classifier_tensorflow_embedding”
sample NLU_data.md [I am just giving an example]
intent:inform
. .
Example Scenerio User - Hi Bot - Welcome to BookS 1. Add Book 2. Remove Book 3. Book Issue 4. Book Info User - Book Issue Bot - Sorry i didn’t understand you # Fallback or # Circuit Tripped error
Action endpoint error DEBUG:rasa_core_sdk.forms:Validating user input ‘{‘intent’: {‘name’: ‘inform’, ‘confidence’: 0.685754120349884}, ‘entities’: [], ‘intent_ranking’: [{‘name’: ‘inform’, ‘confidence’: 0.685754120349884}, {‘name’: ‘Book Issue’, ‘confidence’: 0.6697322130203247}, {‘name’: ‘Book Info’, ‘confidence’: 0.10917111486196518}, {‘name’: ‘stop’, ‘confidence’: 0.09826930612325668}, {‘name’: ‘thankyou’, ‘confidence’: 0.0013971030712127686}, {‘name’: ‘deny’, ‘confidence’: 0.0}, {‘name’: ‘greet’, ‘confidence’: 0.0}, {‘name’: ‘filewatch_issue’, ‘confidence’: 0.0}, {‘name’: ‘access’, ‘confidence’: 0.0}, {‘name’: ‘cancel’, ‘confidence’: 0.0}], ‘text’: Book Issue’}’
As you can see entities: [] is empty but if i retrain the model, Book Issue will work but Add Book or Remove Book May not work. So i need train the model and check if every flow(Add Book/Remove Book/Book Issue…etc) is working which is very hectic. I am least confident on the model since i cant ask my client to retrain the model
I Hope you got the gist of what i am trying to convey and please help me out here