Handling Names and Destination Intent in Tensorflow pipeline

Hello all,

I am trying out a flow where bot asks question:

  1. May I know your name?

  2. May I know your destination?

I am using Tensorflow pipeline for Intent classification. The model is not able to identify the intents as the names and destination values are very random in real life.

How to crack this?

Using spacy ner_crf pipeline, it can identify entities for names and destination based on training, but to identify intents will it be better than tensorflow intent classification model? Please share your thoughts regarding these?

Thanks!!

hey @Rajskc you can use Lookup table

https://rasa.com/docs/nlu/master/dataformat/#lookup-tables

tensorflow is usually preferred best for intent classification you can go ahead with that… as compared to spacy the training data is also less

You can use this config for intent + entity extraction -

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” intent_tokenization_flag: true intent_split_symbol: “+”

for more info follow this awesome link -

Tuning Your NLU Model

Do you mean create a list of different names and location? What if someone enters a name outside the list. The model might not recognize right.

I am using the pipeline similar to this. My problem is the classifier: intent_classifier_tensorflow_embedding is not able to generalize well for names and location as an intent.