SAGE Intent Recognition

Independent package for intent recognition and classification in conversational AI systems

🎯 Overview

sage-intent provides flexible intent recognition capabilities for conversational AI:

• Keyword-Based Recognition: Fast, rule-based intent matching
• LLM-Based Recognition: Semantic intent understanding with LLMs
• Hybrid Classification: Combine multiple recognizers
• Extensible Architecture: Easy to add custom recognizers

📦 Installation

# Basic installation (keyword-based only)
pip install isage-intent

# With LLM support
pip install isage-intent[llm]

# Development installation
pip install isage-intent[dev]

🚀 Quick Start

Keyword-Based Recognition

from sage_intent import KeywordIntentRecognizer, IntentCatalog

# Create intent catalog
catalog = IntentCatalog()
catalog.add_intent(
    name="search",
    keywords=["find", "search", "look for", "query"],
    description="Search for information"
)
catalog.add_intent(
    name="greeting",
    keywords=["hello", "hi", "hey"],
    description="Greet the user"
)

# Create recognizer
recognizer = KeywordIntentRecognizer(catalog)

# Recognize intent
intent = recognizer.recognize("Can you help me search for papers?")
print(intent.name)  # "search"
print(intent.confidence)  # 0.85

LLM-Based Recognition

from sage_intent import LLMIntentRecognizer, IntentCatalog

# Create catalog with descriptions
catalog = IntentCatalog()
catalog.add_intent(
    name="data_analysis",
    description="Analyze data, generate visualizations, compute statistics"
)
catalog.add_intent(
    name="code_generation",
    description="Write code, create functions, implement algorithms"
)

# Create LLM recognizer
recognizer = LLMIntentRecognizer(
    catalog=catalog,
    llm_client=your_llm_client
)

# Recognize with semantic understanding
intent = recognizer.recognize(
    "I need a function to calculate the mean and standard deviation"
)
print(intent.name)  # "code_generation"

Hybrid Classifier

from sage_intent import IntentClassifier, KeywordIntentRecognizer, LLMIntentRecognizer

# Create classifier with multiple recognizers
classifier = IntentClassifier(
    recognizers=[
        KeywordIntentRecognizer(catalog),
        LLMIntentRecognizer(catalog, llm_client)
    ],
    strategy="vote"  # or "confidence", "cascade"
)

# Classify with combined approach
intent = classifier.classify("Find research papers about transformers")

📚 Key Components

1. Intent Catalog (catalog.py)

Manages intent definitions:

• Intent registration with keywords and descriptions
• Hierarchical intent organization
• Intent metadata and examples

2. Keyword Recognizer (keyword_recognizer.py)

Fast rule-based matching:

• Multiple keyword patterns per intent
• Fuzzy matching support
• Priority-based disambiguation

3. LLM Recognizer (llm_recognizer.py)

Semantic understanding with LLMs:

• Zero-shot intent classification
• Few-shot with examples
• Confidence scoring

4. Classifier (classifier.py)

Combines multiple recognizers:

• Voting strategies
• Confidence-based selection
• Cascade fallback

5. Factory (factory.py)

Easy recognizer creation:

• Pre-configured recognizers
• Custom recognizer registration
• Dynamic loading

🔧 Architecture

sage_intent/
├── base.py                  # Base classes and protocols
├── types.py                 # Common types
├── catalog.py               # Intent catalog management
├── keyword_recognizer.py    # Keyword-based recognition
├── llm_recognizer.py        # LLM-based recognition
├── classifier.py            # Multi-recognizer classification
├── factory.py               # Recognizer factory
└── __init__.py             # Public API exports

🎓 Use Cases

1. Chatbots: Route user queries to appropriate handlers
2. Voice Assistants: Understand user commands
3. Customer Support: Classify support tickets
4. Search Systems: Detect search intent for better results
5. Agent Systems: Determine agent actions based on user intent

🔗 Integration with SAGE

This package is part of the SAGE ecosystem but can be used independently:

# Standalone usage
from sage_intent import KeywordIntentRecognizer, IntentCatalog

# With SAGE agentic (optional)
from sage_agentic import Agent
from sage_intent import IntentClassifier

agent = Agent()
classifier = IntentClassifier(catalog)

def process_query(query):
    intent = classifier.classify(query)
    return agent.execute(intent)

📖 Documentation

• Repository: https://github.com/intellistream/sage-intent
• SAGE Documentation: https://intellistream.github.io/SAGE-Pub/
• Issues: https://github.com/intellistream/sage-intent/issues

🤝 Contributing

Contributions are welcome! Please feel free to submit a Pull Request.

📄 License

MIT License - see LICENSE file for details.

🙏 Acknowledgments

Originally part of the SAGE framework, now maintained as an independent package for broader community use.

📧 Contact

• Team: IntelliStream Team
• Email: shuhao_zhang@hust.edu.cn
• GitHub: https://github.com/intellistream