Spaces:
Sleeping
Sleeping
File size: 10,687 Bytes
d36787a e3fee32 d36787a e3fee32 d36787a e3fee32 d36787a |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 |
"""
Intent Classification Service
Determines user intent to make the chatbot behave naturally
"""
import re
from typing import Dict, Any
class IntentClassifier:
"""Classifies user intents to determine appropriate bot response"""
# Keywords that indicate pipeline creation intent
PIPELINE_KEYWORDS = [
"extract", "summarize", "translate", "classify", "detect",
"analyze", "process", "generate", "create pipeline", "build pipeline",
"run", "execute", "perform", "do", "get", "find", "identify",
"table", "text", "image", "signature", "stamp", "ner", "entities"
]
# Casual chat patterns
CASUAL_PATTERNS = [
r"^(hi|hello|hey|greetings|good morning|good afternoon|good evening)",
r"^(how are you|what's up|wassup)",
r"^(thanks|thank you|appreciate)",
r"^(bye|goodbye|see you|later)",
r"^(ok|okay|cool|nice|great|awesome)",
r"^(what can you do|what do you do|help|capabilities)",
r"^(who are you|what are you)"
]
# Question patterns that need informational response
QUESTION_PATTERNS = [
r"^(what|how|why|when|where|who|can you|do you|are you|is it)",
r"(help|explain|tell me|show me)"
]
@staticmethod
def classify_intent(user_message: str) -> Dict[str, Any]:
"""
Classify user intent from their message
Returns:
{
"intent": "casual_chat" | "question" | "pipeline_request" | "approval" | "rejection",
"confidence": float (0-1),
"requires_pipeline": bool,
"suggested_response_type": "friendly" | "informational" | "pipeline_generation"
}
"""
message = user_message.strip().lower()
# Empty message
if not message:
return {
"intent": "casual_chat",
"confidence": 1.0,
"requires_pipeline": False,
"suggested_response_type": "friendly"
}
# Approval/Rejection patterns (for pipeline confirmation)
if message in ["approve", "yes", "y", "ok", "okay", "proceed", "go ahead", "do it"]:
return {
"intent": "approval",
"confidence": 1.0,
"requires_pipeline": False,
"suggested_response_type": "execute_pipeline"
}
if message in ["reject", "no", "n", "cancel", "stop", "don't"]:
return {
"intent": "rejection",
"confidence": 1.0,
"requires_pipeline": False,
"suggested_response_type": "friendly"
}
# Casual chat patterns
for pattern in IntentClassifier.CASUAL_PATTERNS:
if re.search(pattern, message, re.IGNORECASE):
return {
"intent": "casual_chat",
"confidence": 0.9,
"requires_pipeline": False,
"suggested_response_type": "friendly"
}
# Question patterns (informational) - CHECK THIS FIRST!
is_question = False
for pattern in IntentClassifier.QUESTION_PATTERNS:
if re.search(pattern, message, re.IGNORECASE):
is_question = True
break
# Check for pipeline keywords
pipeline_keyword_count = sum(
1 for keyword in IntentClassifier.PIPELINE_KEYWORDS
if keyword in message
)
# IMPORTANT: Questions take priority over pipeline keywords!
# "tell me about X" is a question, not a pipeline request
if is_question:
# If it's clearly asking about the system itself, it's informational
if any(term in message for term in ["masterllm", "you", "this", "system", "bot", "chatbot", "assistant"]):
return {
"intent": "question",
"confidence": 0.95,
"requires_pipeline": False,
"suggested_response_type": "informational"
}
# If it's a question but with strong pipeline keywords, could be pipeline request
# But only if the keywords are very explicit (e.g., "extract text from page 5")
elif pipeline_keyword_count >= 2 and not message.startswith("tell me"):
return {
"intent": "pipeline_request",
"confidence": 0.7,
"requires_pipeline": True,
"suggested_response_type": "pipeline_generation",
"keyword_matches": pipeline_keyword_count
}
# Otherwise, it's just a question
else:
return {
"intent": "question",
"confidence": 0.85,
"requires_pipeline": False,
"suggested_response_type": "informational"
}
# If has pipeline keywords but is NOT a question, then pipeline request
if pipeline_keyword_count > 0:
return {
"intent": "pipeline_request",
"confidence": min(0.6 + (pipeline_keyword_count * 0.1), 1.0),
"requires_pipeline": True,
"suggested_response_type": "pipeline_generation",
"keyword_matches": pipeline_keyword_count
}
# Default: treat as casual if short, otherwise might be pipeline request
if len(message.split()) < 3:
return {
"intent": "casual_chat",
"confidence": 0.6,
"requires_pipeline": False,
"suggested_response_type": "friendly"
}
# Longer messages without clear intent - ask for clarification
return {
"intent": "unclear",
"confidence": 0.4,
"requires_pipeline": False,
"suggested_response_type": "clarification"
}
@staticmethod
def get_friendly_response(intent: str, user_message: str = "") -> str:
"""Generate friendly chatbot responses for non-pipeline intents"""
message_lower = user_message.lower().strip()
# Greetings
if re.search(r"^(hi|hello|hey)", message_lower):
return "Hello! 👋 I'm MasterLLM, your AI document processing assistant. Upload a document and tell me what you'd like to do with it!"
# How are you
if "how are you" in message_lower:
return "I'm doing great, thank you! 🤖 Ready to help you process documents. Upload a file to get started!"
# Thanks
if re.search(r"^(thanks|thank you)", message_lower):
return "You're welcome! 😊 Let me know if you need anything else!"
# Goodbye
if re.search(r"^(bye|goodbye)", message_lower):
return "Goodbye! 👋 Feel free to come back anytime you need document processing help!"
# Capabilities question
if "what can you do" in message_lower or "capabilities" in message_lower:
return """I can help you with various document processing tasks:
📄 **Text Operations:**
- Extract text from PDFs and images
- Summarize documents
- Translate to different languages
- Classify text content
- Extract named entities (NER)
📊 **Table Operations:**
- Extract tables from documents
- Analyze tabular data
🖼️ **Image Operations:**
- Describe images
- Detect signatures
- Detect stamps
🔧 **How to use:**
1. Upload a document (PDF or image)
2. Tell me what you want to do (e.g., "extract text and summarize")
3. I'll create a pipeline for you to approve
4. Watch the magic happen! ✨"""
# Who are you / What are you / Tell me about MasterLLM
if any(phrase in message_lower for phrase in ["who are you", "what are you", "about masterllm", "about you", "tell me about"]):
return """**About MasterLLM** 🤖
I'm MasterLLM, an intelligent AI-powered document processing orchestrator that helps you automate complex document workflows.
**What I Do:**
- 🔍 **Analyze** your requests and understand what you need
- 🛠️ **Create** custom processing pipelines automatically
- ⚡ **Execute** multi-step document workflows
- 📊 **Deliver** structured results
**My Capabilities:**
- Extract text from PDFs and images (OCR)
- Summarize long documents
- Translate content to different languages
- Classify documents
- Extract tables and structured data
- Detect signatures and stamps
- Describe images
- Extract named entities (people, places, organizations)
**How I'm Different:**
I'm not just a single tool - I'm a **smart orchestrator** that:
1. Understands your natural language requests
2. Determines the best sequence of operations
3. Uses powerful AI models (AWS Bedrock Claude & Google Gemini)
4. Combines multiple tools into efficient pipelines
5. Executes everything seamlessly
**My Role:**
Think of me as your AI assistant for document processing. I have a natural conversation with you, understand what you need, propose a plan, and execute it - all while keeping you informed every step of the way!
Want to see me in action? Just upload a document and tell me what you'd like to do with it! 🚀"""
# Help
if message_lower in ["help", "?"] or "help me" in message_lower:
return """Here's how to use me:
1️⃣ **Upload Document**: Click the upload button and select a PDF or image
2️⃣ **Describe Task**: Tell me what you want (e.g., "extract text from pages 1-5 and summarize")
3️⃣ **Review Pipeline**: I'll show you the processing plan
4️⃣ **Approve**: Type 'approve' to execute or 'reject' to cancel
5️⃣ **Get Results**: Watch real-time progress and get your results!
**Example requests:**
- "extract text and summarize"
- "get tables from pages 2-4"
- "translate to Spanish"
- "detect signatures and stamps"
Need anything else?"""
# Unclear intent
if intent == "unclear":
return "I'm not sure what you'd like me to do. Could you please:\n- Upload a document first, or\n- Tell me what processing task you need (e.g., 'extract text', 'summarize', 'translate')\n\nType 'help' to see what I can do!"
# Default friendly response
return "I'm here to help! Upload a document and tell me what you'd like to do with it. Type 'help' if you need examples! 😊"
# Singleton instance
intent_classifier = IntentClassifier()
|