Watson Conversation supports many Intents with same common words like 'Yes'?

Question

Watson Conversation supports many Intents with same word? How to avoid wrong confidence association?

Example: Did you see that video http://....? Condition: Intent #see with values: "I see", "I saw", "Yes", etc..

Did you read that blog post http://....? Condition: Intent #read with values: "I read", "Yes", etc..

I link a question with another using "continue conversation" (the + icon at right of each dialog box). What I'm expecting is: If first dialog (see video) is already visited in the conversation and I'm using "continue conversation" feature, the second 'Yes' should answer the second question.

But instead of that every 'Yes' is trying to back the conversation to the root dialog (near the conversation start) because it is always pointing to the same Intent.

So, reuse the same common word in many Intents could cause a crazy conversation, right? The dialog flow and dialogs already visited is not taking in consideration to change de confidence rating, right?

What is the recommendation here? Create a separated Intent to each common word like 'Yes', 'No', 'Sure'?

So each dialog condition logic should include N intents? Example: #see OR #yes OR #sure

Answer 1

In general it is not a good idea to use the same example for two different intents - this basically confuses the classifier as it is now not sure what to output. The classifier - at least for the time being - is not using the conversation context in the classification.

So the intents should be defined as explicitly as possible. In this particular use case I would recommend to use dedicated intents for #yes and #no. You can then reuse these intents at different places in the conversation.