Slot Filling Chatbot开放式提问的意图设计方案咨询

Hey there! Let’s tackle your slot-filling chatbot design question—focused on making early conversations feel natural with open-ended questions, plus figuring out intent structure and context for your three scenarios.

The key here is to avoid making users feel like they’re filling out a form. Instead, design intents that guide users to share information organically, rather than forcing them to answer rigid slot-specific questions.

1. Intent Structure: To Split or Not Split Scenarios 1, 2, 3? #

This depends entirely on how distinct your three scenarios are:

• If scenarios are completely separate tasks (e.g., Scenario 1 = booking a hotel, Scenario 2 = ordering groceries, Scenario 3 = scheduling a doctor’s visit):
  • Yes, create independent intents for each (e.g., BookHotel, OrderGroceries, ScheduleAppointment). Each intent will have its own set of slots and open-ended prompt logic tailored to that task. For example, the BookHotel intent might open with, “What kind of trip are you planning? A cozy weekend getaway or a business trip where you need something close to downtown?”
• If scenarios are stages of the same task (e.g., Scenario 1 = initial food order request, Scenario 2 = specifying dietary preferences, Scenario 3 = confirming delivery details):
  • No need for separate intents. Use a single main intent (e.g., OrderFood) paired with context tracking to manage which stage the user is in. This keeps the conversation fluid, as the bot can reference previous info to frame open-ended questions naturally.

2. Designing Open-Ended Flows for Each Scenario #

Whichever intent structure you choose, wrap slot requests in conversational, open-ended prompts instead of direct demands:

• Bad example (rigid slot ask): “What size? What toppings?”
• Good example (open-ended, multi-slot prompt): “Tell me about your perfect pizza—do you go for a classic pepperoni, or something more adventurous like veggie? And do you usually get a small for yourself, or a large to share?”
• Bonus: Let the bot extract multiple slots from one user response (e.g., if they say “I want a large veggie pizza with extra cheese,” the bot can pull size, toppings, and extra slots all at once).

3. Context Management for Scenarios 2 & 3 #

If Scenarios 2 and 3 are follow-up stages of a conversation (e.g., after Scenario 1’s initial request), you absolutely need to add context flow. Here’s why and how:

• Why context matters: It prevents repetitive questions (if the user already mentioned they’re allergic to nuts, the bot won’t ask about dietary restrictions again) and lets the bot transition smoothly between stages.
• How to implement it: Use context variables (e.g., user_order_context) to store all extracted slots and conversation history. For each scenario stage, check what info is already known, then frame an open-ended prompt for the missing details.
  • Example flow:
    1. Scenario 1 (initial intent): Bot asks, “What kind of meal are you in the mood for today?”
    2. User replies: “I want sushi.”
    3. Scenario 2 (context-aware follow-up): Bot says, “Sushi’s a great pick! Do you have any favorite rolls, or dietary needs I should know about?”
    4. User replies: “I love salmon rolls and can’t eat avocado.”
    5. Scenario 3 (final confirmation): Bot says, “Got it—salmon rolls, no avocado. Would you like that delivered to your usual address, or a new one?”

4. Quick Pro Tips for More Natural Conversations #

• Embrace partial answers: If a user only provides part of the needed info, follow up with an open-ended prompt instead of a yes/no or rigid question. For example, if they say “I want sushi,” reply with, “Nice choice! Do you have a go-to roll, or want me to suggest some popular options?”
• Add conversational flair: Throw in small, natural responses like “Got it!” or “That sounds delicious!” between questions to avoid feeling robotic.
• Handle detours gracefully: If the user goes off-topic (e.g., “It’s raining today!”), acknowledge it first (“Ugh, rain can be such a downer!”) then gently steer back to the task: “Anyway, back to your sushi order—did you want to add any sides like edamame?”

内容的提问来源于stack exchange，提问作者Mikul