Wrong intent prediction


I have got 600 intents in my bot. I am using tensorflow_embedding as pipeline. When the intent count was 250, bot was working fine but as the intents increased to 600, it has started predicting wrong intents.

Is it a overfitting problem ? Can anyone suggest pipeline configuration that might help or any other that can help in proper intent prediction

Hey @vinayver198. It’s a bit hard to say where the issue hides not knowing how your data looks like. How many examples do you have for each intent? Also, before jumping into modifying pipeline configuration, I would suggest inspecting the intents themselves - maybe you have overlaping intents in your training data or intents where an entity would do a better job when designing a bot?


I have 20 questions for every intent and framing entities for every question is a bit hard. Can you suggest something. Or else if possible can I add more features to NN by exctracting it externally to improve its performance ?

Why are such questions unanswered :frowning: