Intent miss classification with 3 classes

I am using following version- rasa-core == 0.11.7 rasa-core-sdk == 0.11.5 rasa-nlu == 0.13.4

I am training a model with three intent.

    • intent_one with 6 utterances -(greetings like hi,hello, hiii, howdy)
    • intent_two with 6 utterances- ( bye, thanks, thank you, see u )
    • intent_three with 7000 utterances- ( observation - on which i am doing my further processing )

With Tensor flow embedding i am not able to classify "hello ", “hi” as greetings intent. Although spacy is doing fine in classification. Any possible reason for this miss-classification? Attaching an error image .tf_error

Hey @kamlesh. Since tensorflow_embedding model doesn’t use pretrained word vectors, it needs more data to learn the embeddings well. However, in your case I think the problem is because of highly imbalanced classes. You defeinitely need more examples for the first two intents for the tf model to perform better.