Bluemix Conversation, advancing the conversation

Question

I have two conversation nodes defined. The first is triggered on conversation_start and asks the users's name. The second stores the user name from input.text in a context variable and asks if the user wants to take a brief survey or do something else. The second one never executes. What should the trigger be on the second one in order to advance the conversation? There is no intent or condition set yet. Basically, I just want to follow a string-of-pearls pattern where every question gets asked and let the user opt out of some of the questions.

Answer 1

there are two main ways to achieve this. The one I recommend is to create an intent for each possible way. For example, you could create an intent called #affirmative, with values like "Yes", "Yeah", "Sure", "Ok", etc. All you have to do is to create the intent and then use it in the condition field.

But you can also check the user's input directly, using input.text. For example, you could use input.text.toLowerCase() == 'yes' as the condition. Have in mind that this should only be used if you are controlling the input and only allowing a few possible values. If not, use an intent and let Watson Conversation handle the input parsing for you.

Another really cool feature that Watson Conversation provides that can benefit is you is to have a node with a condition like intents[0].confidence < 0.6 and a response like "Sorry, I didn't get that". That way you make sure your flow will only trigger a condition with a high confidence for the intent.

Hope it helps!

Answer 2

You can use the jump to function to route the system to arbitrary dialog_node - this is great for creating directed dialogs. So whenever you want to continue with the next question (user answered the previous question), you can use jump to from the node processing the answer to the previous question to ask the user another question (jump to the body part of the next dialog node asking new question).

Now for "get name" use case you want the second node gathering the name from the user input hanging off the conversation_start node. If the conversation encounters a node that has children, the processing of the next dialog round will start at the first child of this dialog node - this way the contextual dialogs are defined in the conversation.

Answer 3

Bluemix Conversation can be thought of as a state machine. Each transition is diagrammed in the tooling, but each node can only be triggered by a condition. So that if you're on a node asking a question, and the answer that comes back does not trigger one of the child nodes, the state will not transition to a child. Answers entered into a node are processed by Watson, who provides "intents", which are essentially keywords found in the sentence he just analyzed. So one way to advance the conversation is to match named "intents" using the # prefix to the trigger of in a child node. Another way it to match named "entities" using the @ prefex, although it is less clear how Watson builds entities. The third way is to match context variables using the $ prefix and simple conditional operators such as ==. Context variables are entirely managed by you, the developer. You are allowed to set one context variable at each node. And then you can use it to trigger a child node. This appears to be a good approach to managing the type of conversation I want to implement. The fourth way is to use the previous parent node's input value using along with simple conditional operators.