如何在Alexa与Dialogflow同一意图中获取多组动态及未知文本

Hey there! I’ve dealt with similar setup challenges when building voice experiences with Alexa and Dialogflow, so let’s break this down step by step for your two questions:

The key here is tracking conversation state so your assistant knows what information it still needs to gather. Here’s how to implement this for both platforms:

For Dialogflow #

• Use Contexts to keep track of what input you’re expecting next. Create a context (e.g., awaiting_content_details) with a lifespan of 2-3 rounds (enough to cover the user’s response).
• Optimize Training Phrases to include both single and multi-input scenarios, like:
  • "I want to add content: title is [title], description is [description]"
  • "First the title: [title], then the description is [description]"
  • "Title is [title], I’ll tell you the description later"
• Build Fulfillment Logic to check which fields are missing, prompt the user, and retain the context. Example Node.js code:

function handleContentIntent(agent) {
  const title = agent.parameters.title;
  const description = agent.parameters.description;
  const currentContext = agent.context.get('awaiting_content_details');

  if (!title) {
    agent.context.set({ name: 'awaiting_content_details', lifespan: 2 });
    agent.add("Could you tell me the title of your content?");
  } else if (!description) {
    agent.context.set({ name: 'awaiting_content_details', lifespan: 2 });
    agent.add(`Got it, title is ${title}. Now can you share the description?`);
  } else {
    agent.add(`Perfect! I've saved your content: Title - ${title}, Description - ${description}`);
  }
}

• Don’t forget to add awaiting_content_details as an Input Context for your intent, so Dialogflow routes follow-up responses back to this intent.

For Alexa #

• Use Session Attributes to store information collected so far during the conversation.
• Implement Intent Handling that checks the session state to know what input to request next. Example Node.js code:

const ContentIntentHandler = {
  canHandle(handlerInput) {
    return Alexa.getRequestType(handlerInput.requestEnvelope) === 'IntentRequest'
      && Alexa.getIntentName(handlerInput.requestEnvelope) === 'ContentIntent';
  },
  handle(handlerInput) {
    const sessionAttributes = handlerInput.attributesManager.getSessionAttributes();
    const userInput = handlerInput.requestEnvelope.request.intent.slots.FreeText.value;

    if (!sessionAttributes.title) {
      sessionAttributes.title = userInput;
      handlerInput.attributesManager.setSessionAttributes(sessionAttributes);
      return handlerInput.responseBuilder
        .speak(`Got the title: ${userInput}. Now please tell me the description.`)
        .reprompt("Could you share the description for your content?")
        .getResponse();
    } else if (!sessionAttributes.description) {
      sessionAttributes.description = userInput;
      return handlerInput.responseBuilder
        .speak(`Great! I've recorded your content: Title is ${sessionAttributes.title}, Description is ${userInput}`)
        .getResponse();
    }
  }
};

• Use AMAZON.SearchQuery as the slot type (equivalent to Dialogflow’s @sys.any) to capture arbitrary user text.

This relies on guiding the conversation and training your model to infer which input corresponds to which field:

• Leverage Step-by-Step Prompting: Instead of letting the user ramble, explicitly ask for one field at a time. For example:

  1. First prompt: "What’s the title of your content?"
  2. After receiving the title: "Thanks! Now can you describe the content?"
     This eliminates ambiguity because each user response is tied to a specific request.

• Optimize Training Phrases for Ambiguous Inputs: Add examples where the user provides both fields without labels, like:

  • "[title] [description]"
  • "[title], it’s about [description]"
    Dialogflow’s ML will learn to map the first part to the title slot and the rest to the description slot over time.

• Use Dialogflow’s Auto Slot Filling: If you’ve defined title and description as separate slots (both using @sys.any), enable auto slot filling. Configure prompts for each missing slot, and Dialogflow will automatically ask for the missing information if the user only provides one field.

• Lightweight Semantic Checks (Optional): In fulfillment, you can add simple logic to guess which field the user provided (e.g., short text = title, longer text = description), but this is a fallback—context-guided prompting is much more reliable.

内容的提问来源于stack exchange，提问作者Ruby