How much data do I need to train the Intent Detector?
A minimum of 5 examples for each intent is required to train the Intent Detector. However, the more examples taught, the smarter and more accurate the assistant.
Balance should also be considered. For example, if there are 300 training examples for Intent A and only 5 for Intent B, the assistant may be more inclined to think that the user's input is consistent with Intent A than with Intent B. Imagine what a person would know if we taught him for a long time to express Intent A in many different ways, and only gave him a few examples of Intent B in general.
Usually, with a few dozen examples for each intent, the virtual assistant works pretty accurately.
The virtual assistant knows more if it is given as many examples of intents as possible. These examples should have different phrases and word combinations. It also works better when the prepared intent examples consist of complete questions and sentences rather than a single word.