An interaction model is a key component of any conversational AI system. It defines how users will interact with the system and how the system will respond. The main components of an interaction model include:
Intents
Intents represent the purpose or goal behind a user’s input. They categorize the user’s intent and help the system determine how to respond appropriately. Common intents include asking a question, making a purchase, checking account balance, etc. Defining clear intents is crucial for an effective interaction model.
Examples of intents
- Check balance
- Make payment
- Order food
- Book appointment
Entities
Entities represent key information in the user’s input that is relevant to their intent. This includes things like names, locations, amounts, etc. Extracting entities from the input allows the system to take action on specific details. For example, for a payment intent, entities would include payment amount, date, recipient name, etc.
Examples of entities
- Date
- Time
- Location
- Person name
User Inputs
User inputs represent the different ways users may phrase requests or queries to the system. Defining expected sample user inputs for each intent helps train the machine learning model to recognize intents accurately. Variations in phrasing, word order and sentence structure should be included.
Examples of user inputs
Intent | Sample User Inputs |
---|---|
Check balance | How much money do I have in my account? |
What is my account balance right now? | |
Can you tell me my current balance please? | |
Make payment | I need to pay John $50 |
Transfer $100 to my mom for her birthday |
Dialog Flow
The dialog flow defines the conversation structure and models how the interaction will logically flow between the user and the assistant. This includes things like:
- How the system will prompt for needed info
- How the system will handle questions, interrupts, switching conversation topics, etc.
- When the system will trigger business logic or external APIs
- How the system will route the user through various paths depending on context
Defining these dialog flows allows you to create logical, natural conversations that meet the user’s needs.
Example dialog flow
# | Actor | Utterance |
---|---|---|
1 | System | Hi, how can I help you today? |
2 | User | I’d like to check my account balance please |
3 | System | Okay, let me pull up your account. What is your account number? |
4 | User | It’s 123456789 |
5 | System | Thank you. I’ve pulled up your account, and your current balance is $3,582.67. Is there anything else I can help you with today? |
Responses
Responses represent the system’s replies to the user for each intent. Variations in tone, wording, and phrasing should be included to make conversations more natural. Responses can include:
- Confirmations like “Okay, let me check that for you”
- Information like account balances, order statuses, etc.
- Requests for needed info to complete the user’s request
- Questions to clarify user intent if needed
Well-defined responses are crucial for conversing naturally and providing users with the information they need.
Example responses
Intent | Sample Responses |
---|---|
Check balance | Got it, let me pull up your current balance. |
Let me take a look at your account balance for you. | |
No problem, I can definitely check your balance. Can you confirm your account number for me? | |
Make payment | Okay, I can help make that payment. Who would you like to pay and how much? |
Sure thing, just confirm the payment details and I’ll submit that payment for you. |
Integration
Integration enables connecting the conversational AI with other systems like databases, APIs, business logic, etc. This allows the assistant to pull real data and take actions on behalf of the user. Some examples include:
- Looking up account balances from a CRM
- Submitting payment transactions via payment gateways
- Scheduling calendar appointments in a calendar app
- Retrieving inventory availability from an ERP system
Smooth integration creates seamless conversations that provide true value to users.
Common integration methods
Method | Description |
---|---|
APIs | Integration via REST or web APIs to connect to external systems and data |
Webhooks | Triggers that notify external systems of events like new messages |
Database access | Direct access to databases to retrieve and write data |
Embedded code | Custom code embedded in the assistant to execute logic |
Conclusion
An effective interaction model requires clear intents, entities, user inputs, dialog flows, responses, and integration capabilities. Defining these components creates a robust framework for natural conversations that meet users’ needs. With a well-designed interaction model, conversational AI systems can deliver truly intelligent and intuitive assistance to users.