Logo of AI Product Engineer
AI Product Engineer
Tutorial Image: LangGraph Tutorial: Implementing Message Classification Nodes - Unit 1.3 Exercise 2

LangGraph Tutorial: Implementing Message Classification Nodes - Unit 1.3 Exercise 2

Learn how to implement message classification in LangGraph using dedicated classifier nodes. This tutorial covers content analysis, confidence scoring, and state preservation to create robust systems for intent-based routing and automated decision-making.

🎯 What You'll Learn Today

LangGraph Tutorial: Implementing Message Classification Nodes - Unit 1.3 Exercise 2

This tutorial is also available in Google Colab here or for download here

Joint Initiative: This tutorial is part of a collaboration between AI Product Engineer and the Nebius Academy.

This tutorial demonstrates how to implement a message classification system in LangGraph using dedicated classifier nodes. We'll create a node that can analyze message content and assign both classifications and confidence scores.

Key Concepts Covered

  1. Message Content Analysis
  2. Classification Logic
  3. Confidence Score Assignment
  4. State Preservation
from typing import Annotated, TypedDict
!pip install langchain-core
!pip install langgraph
from langchain_core.messages import BaseMessage, HumanMessage
from langgraph.graph.message import add_messages

Step 1: State Definition for Classification

We define a state structure that supports classification operations and confidence tracking.

class State(TypedDict):
    """State container optimized for message classification.

    This implementation focuses on three key aspects:
    1. Message storage with proper annotations
    2. Classification category assignment
    3. Confidence score tracking

    Attributes:
        messages: List of conversation messages with add_messages annotation
        classification: Current message classification label
        confidence: Classification confidence score (0.0 to 1.0)

    Note:
        The combination of classification and confidence enables
        more sophisticated routing and decision-making logic.
    """

    messages: Annotated[list[BaseMessage], add_messages]
    classification: str
    confidence: float

Why This Matters

Message classification is crucial because it enables

  1. Intent-based conversation routing
  2. Automated response selection
  3. Analytics and monitoring
  4. Quality control through confidence tracking

Step 2: Classifier Node Implementation

We implement the core classification logic with confidence scoring.

def classifier_node(state: State) -> State:
    """Analyze and classify conversation messages with confidence scoring.

    This node implements several key concepts:
    1. Content-based classification
    2. Confidence score assignment
    3. State preservation during updates

    The classification process follows this flow:
    1. Extract last message content
    2. Analyze content patterns
    3. Assign classification and confidence
    4. Preserve existing state

    Args:
        state: Current conversation state with messages

    Returns:
        State: Updated state with classification and confidence

    Example:
        >>> state = {
        ...     "messages": [HumanMessage(content="Hello!")],
        ...     "classification": "",
        ...     "confidence": 0.0
        ... }
        >>> result = classifier_node(state)
        >>> print(f"{result['classification']}: {result['confidence']}")
        "greeting: 0.9"
    """
    # Extract message content for analysis
    message = state["messages"][-1].content

    # Perform classification analysis
    if "hello" in message.lower():
        return {
            "messages": state["messages"],
            "classification": "greeting",
            "confidence": 0.9,
        }
    else:
        return {
            "messages": state["messages"],
            "classification": "unknown",
            "confidence": 0.1,
        }

Debug Tips

  1. Classification Issues:
  • Log message content before classification
  • Verify classification logic paths
  • Monitor confidence score assignments
  1. State Management:
  • Check message preservation
  • Validate classification updates
  • Verify confidence range
  1. Common Errors:
  • IndexError: Empty message list
  • AttributeError: Invalid message format
  • KeyError: Missing state fields

Key Takeaways

  1. Classification Design:
  • Use content-based analysis
  • Assign meaningful confidence scores
  • Preserve existing messages
  1. State Handling:
  • Maintain message integrity
  • Update classifications atomically
  • Track confidence accurately

Common Pitfalls

  1. Not handling edge cases
  2. Inconsistent confidence scoring
  3. Message list mutation
  4. Missing state preservation

Next Steps

  1. Add pattern recognition
  2. Implement multi-label support
  3. Add confidence thresholds
  4. Enable classification history

Example Usage

initial_state = {
    "messages": [HumanMessage(content="Hello!")],
    "classification": "",
    "confidence": 0.0,
}

result = classifier_node(initial_state)

print(f"Classification: {result['classification']}")
print(f"Confidence: {result['confidence']}")

Variations and Extensions

  1. Enhanced Classification:
  • Regular expression patterns
  • Keyword-based scoring Example use case: Complex intent detection
  1. Advanced Confidence:
  • Multi-factor confidence scoring
  • Context-aware confidence Scenario: High-precision classification

Expected Output

Classification: greeting
Confidence: 0.9

Rod Rivera

🇬🇧 Chapter