Let’s say I have a training phrase like “okay” or “i understand” which I have listed under Intent A, Intent B, and Intent C. When validating the data, of course I get a warning like "okay" was found labeled with multiple different intents in the training data. Each annotated message should only appear with one intent. You should fix that conflict
. However, “okay” does make sense to be matched to these intents, provided in what context that they are in. Ultimately, should I expect that “okay” give like 30% probabilities to Intent A, Intent B, and Intent C? Is there any problem really with having duplicate training phrases? Furthermore, should I expect that these will be fixed provided with the context in previous conversation turns?
Shawn,
No, you should have only one intent with these phrases, not three.
You would then uses stories or forms to handle the conversation flow including the affirm
intent that is used in each of the flows. If you take a look at our example helpdesk bot you’ll seen an affirm
intent and then a form in the action code that uses this intent based upon the dialog.
Our financial demo bot also uses an affirm
intent and you will see stories which drive the conversation based on the users use of affirm
.
Greg
I see, thank you for that insight Greg! I see why having only one intent may be ideal. However, what if in Intent A I had training phrases like “okay”, “i understand”, “i see about the donation description” while Intent B had phrases like “okay”, “i understand”, “let’s go with $35”. Here, although Intent A and Intent B both have similar affirmation phrases, they also have mutually exclusive other phrases going with an intent.
Additionally, could you explain the downsides I may see with what I was doing in the beginning. Would I expect the probabilities for Intent A and Intent B be around the same?