Suppose i have an intent “new_connection” which has sentences like “How to apply for new connection”, “New connection application process” etc. Now, if somebody types “Status of new connection”, then the bot assumes its intent as “new_connection” which is incorrect. Such kind of thing happens at many other places where it finds a match for sentences that have some keywords present in the training data but are not correct sentences for that keyword. So, i wanted to access the confidence value for each reply so that I can impose some sort of threshold in order to avoid such situations.
hey @raghibomar, you can set the threshold value of your rasa-nlu component while training, by passing an argument to the rasa_core.train script using --nlu_threshold
hey @raghibomar you can call the Rasa NLU REST api and it will give you the JSON response and from that you can extract the Confidence score easily. 