Mobi Copilot provides comprehensive features for configuring intents. You can use policies, such as a Large Language Model (LLM) or keyword matching, to recognize user intents and attach corresponding operations. This topic describes how to configure intents on the Mobi platform.
What is intent recognition
Intent recognition enables a conversational bot to understand the intent or purpose that a user expresses in natural language. Intent recognition is a typical feature of intelligent assistants. For example, if a user enters "I want to book a meeting room for today" in a conversation, "book a meeting room" is the user's intent. This is the operation the user wants the conversational service to perform.
Features
The Copilot intent settings feature supports the following:
Set intent recognition policies: You can use policies such as a Large Language Model (LLM) or keyword matching.
Attach various operation types: You can attach operations to intents, such as Model Studio applications, HTTP integrations, and function routing.
Immediate effect: After you change and save the intent settings, you can test the new intents in your Copilot application.
Create an intent
1. Create an intent skill
Log on to the Mobi platform and go to the Designer.
In the navigation pane on the left, go to the Intent Settings panel.
Click Create Intent Skill to add a new intent skill.
You can create a maximum of 50 intent skills for each Copilot application.
2. Set basic information for the intent skill
In the intent skill settings, fill in the following basic fields:
Skill Name: The name of the skill. Use a descriptive name.
Skill Type: Determines the type of operation that runs after Copilot matches the intent. The following types are supported:
Function Routing: Allows you to call a function route.
LLM RAG: Allows you to call an integration operation of the type Alibaba Cloud Model Studio Application Integration or HTTP Integration.
Integration Flow: Allows you to call an integration operation of the type Integration Flow.
Skill Description: A detailed description of the skill. A complete description helps the Large Language Model (LLM) better understand the purpose of the skill.
The accuracy and completeness of the skill name and description greatly affect the LLM's ability to understand the intent. Fill in these fields accurately.
3. Attach an operation
After you select a skill type, the platform automatically filters the integration operations or function routes that you can attach. Select the specific operation from the drop-down list.
A shortcut is also available to create new integration operations or function routes.
4. Other settings
1. Keyword matching
In addition to intent recognition using an LLM, you can also use keyword matching.
To use this feature, enable the Match Keywords option and set your keywords.
When keyword matching is enabled, the Copilot service first tries to match user input against keywords. If the input does not match the keywords of any skill, the service then uses the LLM for intent recognition.
2. Sample questions
You can add up to 10 sample questions. This helps the LLM understand the scenarios for the intent skill and improves the accuracy of intent recognition.
5. Save
Click Save in the lower-right corner of the Intent Skill panel to save your changes.
After you save the settings, you can enter text in the Copilot input area to test the new intent skill.
6. Set the default intent
After you save the settings, you can configure a default intent in the Default Intent column of the intent skill list. When an intent skill is set as the default, the intent recognition service uses this skill if it cannot match any other intent skill.