microsoft / Cognitive-LUIS-Windows

Windows (.Net) SDK for the Microsoft Language Understanding Intelligent Service API, part of Congitive Services
www.microsoft.com/cognitive-services/en-us/language-understanding-intelligent-service-luis
Other
134 stars 78 forks source link

Increasing the predicted intents for syntax words #64

Closed mynksthi closed 6 years ago

mynksthi commented 6 years ago

How can i increase the prediction intent for the syntax words like order, ordering ,orders ? I have tried by defining them , in list entity (synonym) also i have tried with the add phrase feature How can increase the prediction for it upto minimum 0.7 ?

Thanks

JPThorne commented 6 years ago

In our experience, entities are a decent way, phrase lists too. How much training have you done? i.e. how many utterances do you have mapped to your intents? How many intents do you have?

mynksthi commented 6 years ago

Thanks for your reply i am on a very initial phase of it utterances defined 14-20 intents 6 entities 8-9

JPThorne commented 6 years ago

20 utterances per intent is not too bad; around 50 to 100 is better, so maybe try upping the number of utterances per intent?

also, the similarity of the utterances can bring the score down. i.e. if your utterances all/some have fairly similar structures we found that LUIS get's 'confused' and scores lower.

we treat a score of below 0.5 as being 'too low' & then handle it in code.

mynksthi commented 6 years ago

Where i should define the similiar words

    • order , orders , ordering or
  1. Shipment , supply , export in utterances , features to add phrase or , in entity (list)synonyms
JPThorne commented 6 years ago

We have mainly done that in phrase lists.