Intent recognition extremely low how to improve?

Rasa Open Source
1

Hi there,

My pipeline is below, and I am getting very low recognition of intents even when my reply is exactly the same as in nlu examples. Even when I type ‘yes’ the affirm intent has 0.54 instead of 1.00. Can someone tell me how to solve this?

- name: SpacyNLP
    model: "en_core_web_md" 
  - name: SpacyTokenizer
  - name: SpacyFeaturizer
    pooling: mean
  - name: SpacyEntityExtractor
  - name: CRFEntityExtractor
  - name: LexicalSyntacticFeaturizer
  - name: CountVectorsFeaturizer
  - name: CountVectorsFeaturizer
  - name: "SklearnIntentClassifier"
    C: [1, 5, 10, 50, 100]
  - name: EntitySynonymMapper
2

Any reason you’re using this particular pipeline?

Do you use any metrics tool like Tensorboard?

3

@ChrisRahme no i am not using Tensorboard or similar. Do you know why the recognition is so low?

4

I highly suggest using Tensorboard (see this post to learn more).

Any reason you’re using this particular pipeline? Why not use DIET?