Question Classifier

Node Overview

Node name: Question Classifier
Supported scenarios: Chatflow Agent (supported), Workflow Agent (supported)

Feature Description

The Question Classifier node classifies input questions or recognizes intent, using the classification result as a condition to select the downstream processing branch. It intelligently analyzes user input and automatically identifies the question type or intent, providing the decision basis for subsequent flow branching.

Node Display Content

1. Basic Information Display

• Icon: node-specific icon
• Name: node name
• Model: displayed the same way as the LLM node
• Class ID and class content: displays the configured classification information
• Note: explanatory text added by the user

2. Action Buttons

• More button: click to display the context menu content
• Run button: click to enter the test run interface

Node Edit Panel

1. Large Language Model Configuration

• Model selection: same as the LLM node's model configuration
• Parameter settings: model parameters can be configured

2. Input Variable Configuration

• Selection method: an input box similar to the System Prompt—type / in the box to bring up variable selection and insert a variable (such as Start/sys.query in the screenshot above), rather than a standalone dropdown selector
• Editing aids: supports character count, copy, and maximize (same as the Prompt editor)

3. Class Configuration

• Class management: multiple classes can be displayed; at least 2 classes are required
• Class deletion: classes can be deleted
• Class editing: configuration works the same as the LLM's System_prompt configuration
• Add class: click to add a new class

4. Advanced Settings

• Additional instruction: supports writing an additional prompt instruction
• Configuration method: works the same as the LLM's System_prompt configuration

5. Memory

• Feature: same as the LLM node's memory feature, providing long-term memory and short-term memory switches

6. Output Variables

• Variable name: class_name
• Variable type: string
• Meaning: the class name

7. Next Step Configuration

• Adding a next node is supported
• Each class corresponds to its own next-step branch: on the canvas, each class has its own output connection point; whichever class is matched, that branch is taken (as in the screenshot above: Class 1 → Parameter Extractor, Class 2 → IF/ELSE)

Execution Logic

The execution logic of the Question Classifier node is implemented by the backend to ensure accurate question classification.

Context Menu Actions

The context menu contains the following options:

1. Change node: change the node type
2. Copy: copy the node
3. Duplicate: duplicate the node content
4. Delete: delete the node
5. Help link: jump to the help documentation
6. Run node: test-run the current node

Use Cases

1. Customer Service Systems

• Question classification: classify user questions into technical support, product inquiries, complaints and suggestions, etc.
• Routing and dispatch: route questions to the appropriate handling team based on the classification result
• Priority ordering: set different handling priorities based on question type

2. Content Management

• Content classification: automatically classify articles, comments, feedback, and other content
• Tag generation: automatically generate relevant tags for content
• Content recommendation: recommend related content based on classification results

3. Business Processes

• Ticket classification: automatically classify ticket types
• Requirement analysis: analyze the types of user requirements
• Flow branching: select different processing flows based on classification results

4. Data Analysis

• Sentiment analysis: analyze the sentiment of text
• Topic identification: identify the main topics of text
• Trend analysis: analyze trends across different categories

Class Configuration

1. Class Design Principles

• Mutual exclusivity: classes should be mutually exclusive, avoiding overlap
• Completeness: classes should cover all possible cases
• Clarity: class names should be clear and specific
• Extensibility: allow new classes to be added later

2. Advanced Settings

• Classification instruction: provide extra classification guidance (written in the Instruction field under Advanced Settings)

Best Practices

1. Class Design

• Hierarchical structure: use a hierarchical classification structure
• Balance: keep the classes balanced
• Maintenance: update and maintain the classification scheme regularly

2. Model Selection

• Accuracy: choose a model with high classification accuracy
• Speed: consider classification processing speed
• Cost: balance accuracy against cost

Configuration Steps

1. Basic Configuration

1. Select an appropriate large language model
2. Configure the input variable
3. Set up the class configuration

2. Class Setup

1. Add class categories
2. Set class names and IDs
3. Configure class descriptions

3. Advanced Configuration

1. Set the advanced instruction
2. Configure memory
3. Set output variables

4. Testing and Validation

1. Test using the test run feature
2. Check classification accuracy
3. Adjust configuration parameters

Considerations

1. Class count: at least 2 classes are required; no more than 20 is recommended
2. Class quality: ensure classification accuracy and consistency
3. Model selection: choose a model suited to classification tasks
4. Data quality: ensure the quality and format of input data
5. Performance considerations: consider the performance and cost of classification processing

FAQ

Q: How do I design a classification scheme?

A: Design a mutually exclusive and complete classification scheme based on business needs, using clear naming.

Q: What if classification is inaccurate?

A: Check the class configuration, adjust the advanced instruction, or switch models.

Q: How many classes can I add?

A: At least 2; no more than 20 is recommended. The exact number depends on business needs.

Q: How do I improve classification accuracy?

A: Provide clear classification instructions and choose an appropriate model.

Q: How is the classification result used in the downstream flow?

A: The classification result can serve as the basis for conditional decisions, selecting different processing branches.