How I encountered this problem:
The user input has several sentences; –For example: Hi Rachel! Do you know where is the nearest hospital? I’m not feeling well.
If I feed all these sentences to my NLU model, the DIET classier will match a wrong intent sometimes and the confidence is below 80%. 
If I only feed one key sentence, for example, Do you know where is the nearest hospital? ,the intent result will be more accurate and the confidence is over 90%. 
I’ve read the bolg:https://rasa.com/blog/how-to-handle-multiple-intents-per-input-using-rasa-nlu-tensorflow-pipeline/ The blog case defines the multiple intents before head.
Is there a way to solve this problem? Or It’s only due to my NLU training data doesn’t cover enough cases?
