Rasa core prediction issue

Whenever i type just random words the rasa core respond it with random intents with >70% confidence. Does anyone has idea about this cause.

which version of rasa_nlu, do you use?

It’s 0.14.6

could you please make an example, so we could reproduce the issue

like when I typed in random words like ‘asdsfsdrrwe’, it pickup an intent with 70-80% accuracy. My Pipeline configuration are below:

  • name: nlp_spacy
  • name: tokenizer_spacy
  • name: intent_entity_featurizer_regex
  • name: intent_featurizer_spacy
  • name: ner_crf
  • name: ner_spacy
  • name: ner_synonyms
  • name: intent_classifier_tensorflow_embedding epochs: 800 intent_tokenization_flag: true intent_split_symbol: “+”

it could happen because asdsfsdrrwe get assigned some random vector by spacy. If you don’t use space, but use CountVectorsFeaturizer it should out put None