Trouble with intent classification

Hey, I’m facing trouble with intent classification using below pipeline: image

I’m using almost 36 intents and my nlu files contains 1460 lines of training data. After training, nlu is not classifying intents with a significant margin and confidence. for example: nlu is classifying an intent only with 30~40% confidence. So is there any way to improve the efficiency of classification.

I’ve also noticed that my model is over fitting the training data and for few intents i don’t have much training date either. Is there any way to improve the results?

Hi kpreetsid,

I suggest adding more training data to the intents that have few example, having a large class imbalance is often bad for classification. You might also want to try different pipelines, see Evaluating Models.