Rasa_nlu returns intent as null for exact copy of text given during training

Rasa Open Source
1

Rasa NLU failed to recognize my intents defined in nlu_train.md (contents copied below)

Only “Hi Buddy” and “Hey Dude” statement returns proper intent as “greet” Other intents are failed to detect. Even tried copying the text as it is in train data.

Is there a problem in my input training data? Please help me.

Shared my training and config pipeline contents below: nlu_train.md:

intent:search_restaurant

  • I want to search Chinese restaurant
  • Suggest some Italian restaurant

intent:greet

  • Hi Buddy!
  • Hey Dude!
  • Welcome to bot world!

Config.yml:

language: “en” pipeline:

  • name: “tokenizer_whitespace”
  • name: “ner_crf”
  • name: “intent_classifier_keyword”
2

I have solved this problem by using below pipeline:

pipeline:

  • name: “tokenizer_whitespace”
  • name: “ner_crf”
  • name: “intent_featurizer_count_vectors”
  • name: “intent_classifier_tensorflow_embedding”

Also the greet intents which detected are part of “intent_classifier_keyword" component. After introducing tensorflow classifier, intents detected perfectly.

3

Yeah the keyword classifier is just a toy example. You should add some more training data as well though for your model to classify things more accurately