An utterance I've labeled isn't being classified correctly - why? How can I fix it?

Question

I've labeled an utterance in LUIS with an intent and entities, but when I run the model on the same utterance, it outputs different intents and entities.  Why?  How can I fix it?

Answer 1

First, make sure your application has been re-trained with all labels you've entered so far.  To do this, click on "Train" in the lower-left hand corner.
If errors are still present after re-training, there are 2 things you can try:  
Give LUIS a "hint" about important words or phrases in your application by adding a feature.  For example, if LUIS is having trouble detecting an entity, add examples of the entity it should detect as a feature, and add
labels that use some of those examples (in this case it's best to use an "exchangeable" feature).  Adding features with multi-word phrases is also very helpful if the order of words is important in your application -- for example, to distinguish
between "I want apples not bananas" and "I want bananas not apples"
Add more examples.  In particularly, when the content of an entity is highly variable and is detected primarily based on the words around it, like the subject of an email, it can take 10 labels or more for the machine learner to determine the pattern.