Bot should not respond on out-of-context data

Prediction of nlu is very high on irrelevant data, how can I handle it.

Ex- INTENT DATA -

intent:whoareyou

  • Can you tell me more about yourself?
  • What are you?
  • what is your personality

USER QUERY- personality is your what

Model prediction score- Near 0.90%

NLU Pipeline-

pipeline:

  • name: ConveRTTokenizer
  • name: ConveRTFeaturizer
  • name: CRFEntityExtractor
  • name: DIETClassifier entity_recognition: False epochs: 300

what is your personality and personality is your what are very close sentences (the order of words is relatively weak signal for ML algorithm), you need to provide these examples in your training data with different intent so that DIETClassifier could be able (not guaranteed) to learn that they are different