Intent Classifier TensorFlow Embedding with SpacyFeaturizer

Rasa Open Source
1

Hi, While testing the results of different configuration options, I found that combining Intent Classifier TensorFlow Embedding with SpacyFeaturizer gives the most logical and flexible results although I couldn’t find any resource recommending or describing using them together. Appreciate if anyone can explain any problem I might be missing. Below is a copy of the pipeline, and training and test data for the greeting intent: pipeline:

  • name: “SpacyNLP”
  • name: “SpacyTokenizer”
  • name: “SpacyFeaturizer” name: “intent_classifier_tensorflow_embedding” Training:

intent:greet

  • hey
  • hello
  • hi
  • Hi Test: hi hello helllo hellllo

Cheers, Mohamed

2

the performance is very dependent on the training data, if you have general words for which spacy embedding are good then it is ok to use this pipeline

1 Like