Watson conversation does not operate as it is mentioned in the documents

Question

I am fairly new to Watson conversation. I found an issue which is matter to my project and appreciate if someone could help me with that.

In Dialog overview document, it is clearly mentioned that Watson will look at the child node to search for further information (preferably through checking the entities).

To test that, I have created a simple environment, two intents as 'find a bar' and 'turn on radio' and one entity as 'music type' with only one value 'Jazz Music'. so if user say 'turn on radio', then the system response will be 'what kind of music do you like?' and the user response is Jazz music and if the user ask find a bar, system will response what music in the bar and user will reply 'Jazz music'.

To implement it, I have created two nodes with a child node each, the parent node recognizes the intent and the child node recognizes the entity (as mentioned in the document). 'find a bar ' is first in the Dialog tree and 'turn on' is the second.

So I tried with the second intent and typed 'turn on'. Intent is recognized correctly and system asked accordingly then I typed Jazz music system response is from the same child (child node of Turn on) but the identified intent is 'Irrelevant'. such response to the application will be confusing if application needs to analyse the response object to identify the intent.

screenshot of the dialog tree and Watson response

Any advise please

Answer 1

This is working as designed and not an issue.

When you make a message call Watson assistant will always try to find the intent and entities.

In your second message you mention Jazz, which is not trained in any of your intents, so it is just mentioning that.

It has no impact on the flow of your design. It's simply a debug window.

Answer 2

If you want to be able to distinguish between both you need to also capture the intent either in separate nodes or you can now use conditional responses per combination of intent and entity.