When I have more huge number of intents e.g. more than 500-800, then how to go about that with so many utters_response and intents & stories. Please help.
When I have more huge number of intents e.g. more than 500-800, then how to go about that with so many utters_response and intents & stories. Please help
That’s a lot of intents… I guess my first piece of advice would be to make sure you really need all of these things as separate intents. Could you have fewer intents but more entities maybe? Or combine some of the intents?
In general you want fewer intents because 1) it’s easier to maintain and test and 2) if you have a large number of intents you need much more training data and you’re likely to see reduced accuracy.
If you have particularly complicated stories, I would recommend upgrading to Rasa 2.0 and using conditional rules to help simplify your assistant as much as possible.
Hi, may I ask what is the ideal range of intent number? Would 100 be too many?
@Crysflair I know this isn’t a satisfying answer, but it depends on the amount of data you have and the patterns in that data. The ideal number of intents is the one that captures all the meaningful, well-represented distinctions in your data.
In general, though, I’ve found that if people have an initial prototype with more than a couple dozen intents it’s often because they’re using intents to store information about entities, which is something you want to avoid. This video goes into more detail about how I would go about deciding what intents to add to a project: Conversational AI with Rasa: Training Data and Rules - YouTube
Thank you for the video! I understand there’s no a one-size-fits-all number of intent, but to start small and add truly necessary intents gradually. As for how to make this decision, nevertheless, requires experience.
Yep, absolutely. If you’re lucky enough to have a conversational designer on your team this is something they can really help with.