0
votes

I'm using Watson Conversation as an FAQ, ie each question that the user asks is expected to NOT relate to the previous one. Expected example:

User: How to book a meeting room? 
Watson: Go to this website, click here, press that....

User: When will I know the results of my application?
Watson: You will receive notice within two weeks if your application is successful.

User: How to upgrade my laptop?
Watson: Log a ticket at this website...

My dialog nodes look like the following, without any child nodes.

welcome
#bookroom
#upgradehardware
#applicationresults
anything_else

But now I am getting the correct intent with high confidence, but the output is always a response from the anything_else node.

User: How to book a meeting room? 
[Intent: #bookroom]
Watson: Sorry I don't understand

User: When will I know the results of my application?
[Intent: #applicationresults]
Watson: Can you please rephrase?

User: How to upgrade my laptop?
[Intent: #upgradehardware]
Watson: I didn't get your meaning.

Help? How am I supposed to treat each node independently?

1
Do the conditions for the dialog nodes match the intent? How is the flow?data_henrik
@data_henrik the condition is exactly the intent. The nodes are like this: if bot recognizes intent, then respond with... , and finally wait for user input, for all of them.Jean
Add <? intents ?> to the anything_else node, and show the results. It is likely it is getting the right intent, but it is below 20% confidence.Simon O'Doherty
@SimonO'Doherty what is "<? intents ?>"? from the json returned by api, the confidence for the correct intent is between 0.8 to 1Jean
That command will show you the full intents+confidences, even if you are not using alternate_intents. If you confidence level is high, then second thing to check is if you stored the intent as "condition" and not an intent. But they fixed that sometime back, so I doubt it is that. One other option is to do a full refresh of your workspace (restart browser). I've seen cases where anything_else got moved, but the UI didn't reflect it. If none of that works, without a reproducible scenario posted not sure how someone could help.Simon O'Doherty

1 Answers

0
votes

Try using Add user Examples in IBM Watson Assistant. Give your custom inputs there (What your customers suppose to ask Training your conversation service ). enter image description here

This may solve your problem.