I'm trying to make a conversational chatbot, so the user inputs are quite wide ranging - beyond just "turn lights on". I want to detect the category of the user intents from their inputs and prepare responses.

I've looked at MS' Luis and api.ai and the intents require a lot of training. Can people suggest other techniques for untrained intent detection?

For example if the user says "Pasta is my favorite dish to cook" then detect "intent preference entity pasta" - then I can gradually build up responses to different categories of inputs.

Perhaps the crowd-sourced intents that wit.ai (facebook) has access to could do this but I'm not sure if all end-users have access to those models.


Browse other questions tagged or ask your own question.