Entity detection in LUIS

1
votes

LUIS does not seem to be good at recognizing entities from utterances that are fed to it. Specifically, if the entity marked in an utterance differs from the sentence that is actually invoked, it is unable to match the entity. For example, If I add an utterance to LUIS:

I want to know more about xyz.

In real world usage, I will replace xyz with some other word, like, say, I want to know more about bot framework. Or, I want to know more about dialogflow

In this case, LUIS is unable to detect 'bot framework' as the entity.

Is there any way around this? Specifically, can i get LUIS to identify the entity that the user has entered, if it does not match with the entity in the sample utterance?

If not, is there any other bot framework that handles this better?

1
botframeworkmicrosoft-cognitiveazure-language-understanding
You would like LUIS to detect "any noun previously unseen by your model", do I get it right? Could you eventually add several examples of whole sentences and expected outcome? - rudolf_franek
How many utterances have you added? Can you post them here so we can look at them? They should have a good degree of difference between them. - DFBerry

1 Answers

3
votes

To make your LUIS app recognize intents and entities better, and improve your LUIS app's performance, you can try:

  • label more utterances and train LUIS for some of those values (bot framework, dialogflow etc)
  • use Phrase lists feature