We are building an app using Watson Conversation but are having a problem handling invalid user responses. Our bot is designed to make hotel bookings and a simplified version of the dialog is:
- Node 1: clear the context; ask “What date do you want to arrive?”
- Node 2: Trigger @sys-date; update context for date; ask “how many nights do you want to stay”
- Node 3: Trigger @sys-number; update context for number of nights; summarise the booking “So you are arriving on [DATE] for [NUMBER OF NIGHTS]”
If the user provides sensible answers then all works but if their response to the check-in date is “blah” it fails so we added a new node (as a sibling of node 2) with an “anything_else” trigger and output “sorry we wanted a date, try again”.
Ideally we would want to loop back to our existing Node 2 so that the trigger @sys-date could catch the user’s revised response but that created an infinite loop so we a child node to “anything_else” that is a copy of Node 2 ie update the context for the date and asks “how many nights…” but now we want to catch the response to that question in node 3 so we can pick up the existing flow so added a jump to node 3. This fails as we end up exiting the dialog branch.
If we use "anything_else" to catch invalid user inputs how do we rejoin the existing dialog flow? At the moment the only solution we have is to duplicate nodes 2 and 3 but in the real dialog branch there are many more nodes so that approach is not practical.
