I am planning to use rasa nlu for intent classification only. I am not building a chat bot. So my aim is use rasa for similarity queries matching. That is to check if query matches any of the configured intent.
So i configured set of intents in rasa and then i try to match it against query.
The issue is even if the text doesnot match the intent i get matching intent with very high confidence. Looks like RASA tries to sum up all confidence values in output to 1.
For example if i have only one intent configured which matches against keyword ‘X’, then all search terms irrespective of what it is will match it as per the current implementation, which looks flawed.