Spaces:
Sleeping
Sleeping
File size: 11,054 Bytes
9a6a5aa |
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 |
from typing import List, Dict, Any
import json
import re
from src.domain.models.conversation import QueryClassification, ConversationContext, Entity
from src.infrastructure.providers.llm_provider import LLMClient
from src.application.services.dynamic_pattern_manager import dynamic_pattern_manager
from config import settings
from config.conversation_config import conversation_config
class QueryClassifier:
"""Classifies user queries to determine if they need new tool data or can be answered from history."""
def __init__(self):
self.llm_client = LLMClient()
async def classify_query(self, user_message: str, conversation_context: ConversationContext) -> QueryClassification:
"""Classify a user query using LLM assistance and rule-based logic."""
# First, try rule-based classification for obvious cases
rule_based_result = self._rule_based_classification(user_message, conversation_context)
if rule_based_result.confidence > 0.8:
return rule_based_result
# Use LLM for more nuanced classification
return await self._llm_classification(user_message, conversation_context, rule_based_result)
def _rule_based_classification(self, user_message: str, conversation_context: ConversationContext) -> QueryClassification:
"""Fast rule-based classification for obvious cases."""
message_lower = user_message.lower()
# Get patterns dynamically from pattern manager
dynamic_pattern_manager.refresh_if_needed()
history_indicators = dynamic_pattern_manager.get_history_patterns()
tool_indicators = dynamic_pattern_manager.get_tool_patterns()
pronoun_patterns = dynamic_pattern_manager.get_pronoun_patterns()
# Check for pronouns without clear antecedents (suggesting reference to history)
has_pronouns = any(bool(re.search(pattern, message_lower)) for pattern in pronoun_patterns)
# Calculate scores
history_score = sum(1 for pattern in history_indicators if re.search(pattern, message_lower))
tool_score = sum(1 for pattern in tool_indicators if re.search(pattern, message_lower))
# Get configuration
config = conversation_config.query_classification
# Check if there's relevant history
has_history = len(conversation_context.turns) > 0
recent_tool_usage = any(turn.tool_used for turn in conversation_context.get_recent_turns(config.recent_turns_check))
# Classification logic using configurable thresholds
if history_score > 0 and has_history:
confidence = min(config.max_confidence, config.min_history_confidence + (history_score * config.confidence_increment))
return QueryClassification(
query_type="history",
confidence=confidence,
reasoning=f"Found {history_score} dynamic history indicators and conversation history exists"
)
if tool_score > 0 or not has_history:
confidence = min(config.max_confidence, config.min_history_confidence + 0.1 + (tool_score * config.confidence_increment))
return QueryClassification(
query_type="tool",
confidence=confidence,
reasoning=f"Found {tool_score} dynamic tool indicators or no conversation history",
needs_tool_data=True
)
if has_pronouns and recent_tool_usage:
return QueryClassification(
query_type="history",
confidence=config.pronoun_confidence,
reasoning="Contains pronouns and recent tool usage suggests reference to history"
)
# Default to tool query with low confidence
return QueryClassification(
query_type="tool",
confidence=config.default_confidence,
reasoning="Uncertain classification, defaulting to tool query",
needs_tool_data=True
)
async def _llm_classification(self, user_message: str, conversation_context: ConversationContext,
rule_based_result: QueryClassification) -> QueryClassification:
"""Use LLM for sophisticated query classification."""
if not settings.HF_TOKEN:
# Fallback to rule-based if no LLM available
return rule_based_result
# Prepare context for LLM
recent_context = self._prepare_context_for_llm(conversation_context)
classification_prompt = self._build_classification_prompt(user_message, recent_context, rule_based_result)
try:
# Get LLM decision
response = await self.llm_client.complete_json(classification_prompt)
return QueryClassification(
query_type=response.get("query_type", rule_based_result.query_type),
confidence=response.get("confidence", rule_based_result.confidence),
reasoning=response.get("reasoning", rule_based_result.reasoning),
referenced_entities=response.get("referenced_entities", []),
needs_tool_data=response.get("needs_tool_data", rule_based_result.needs_tool_data)
)
except Exception as e:
print(f"LLM classification failed: {e}")
# Fallback to rule-based result
return rule_based_result
def _prepare_context_for_llm(self, conversation_context: ConversationContext) -> str:
"""Prepare conversation context for LLM analysis."""
recent_turns = conversation_context.get_recent_turns(5)
if not recent_turns:
return "No previous conversation history."
context_parts = []
for i, turn in enumerate(recent_turns, 1):
turn_summary = f"Turn {i}:\n"
turn_summary += f" User: {turn.user_message}\n"
if turn.tool_used:
turn_summary += f" Tool used: {turn.tool_used}\n"
if turn.tool_params:
turn_summary += f" Parameters: {json.dumps(turn.tool_params, separators=(',', ':'))}\n"
turn_summary += f" Result: {turn.response_summary}\n"
else:
turn_summary += f" Response: {turn.full_response[:100]}...\n"
if turn.extracted_entities:
entities = [f"{e.type}:{e.name}" for e in turn.extracted_entities]
turn_summary += f" Entities: {', '.join(entities)}\n"
context_parts.append(turn_summary)
return "\n".join(context_parts)
def _build_classification_prompt(self, user_message: str, context: str, rule_based_result: QueryClassification) -> str:
"""Build the LLM prompt for query classification."""
return f"""You are a query classification assistant for a Topcoder MCP agent.
Your task is to determine if a user query can be answered from conversation history or needs new external data.
CONVERSATION CONTEXT:
{context}
USER'S NEW MESSAGE: "{user_message}"
RULE-BASED ANALYSIS:
- Preliminary classification: {rule_based_result.query_type}
- Confidence: {rule_based_result.confidence}
- Reasoning: {rule_based_result.reasoning}
CLASSIFICATION TYPES:
1. "history" - Can be answered from conversation context (previous results, mentioned entities, etc.)
2. "tool" - Needs new data from external APIs/tools
3. "chat" - General conversation, greetings, or explanations
ANALYZE THE QUERY FOR:
- References to previous results ("the last challenge", "that user", "it")
- Temporal references ("what you just told me", "from before")
- Pronouns without clear new antecedents
- Requests for new/different/more data vs clarification of existing data
- Entity references that were mentioned in previous conversation
Respond ONLY with a JSON object:
{{
"query_type": "history|tool|chat",
"confidence": 0.0-1.0,
"reasoning": "Brief explanation of your decision",
"referenced_entities": ["entity1", "entity2"],
"needs_tool_data": true/false
}}
Be especially careful to identify when users are asking about previously retrieved data vs requesting new data."""
def extract_entities(self, text: str, previous_entities: List[Entity] = None) -> List[Entity]:
"""Extract entities from text that might be referenced later."""
entities = []
if not conversation_config.entity_extraction.use_dynamic_patterns:
return entities
# Get patterns dynamically from pattern manager
dynamic_pattern_manager.refresh_if_needed()
patterns = dynamic_pattern_manager.get_entity_patterns()
config = conversation_config.entity_extraction
for pattern_name, pattern in patterns.items():
try:
matches = re.finditer(pattern, text, re.IGNORECASE)
for match in matches:
# Get the first non-None group
value = next((group for group in match.groups() if group), None)
if value:
# Determine entity type based on pattern name
entity_type = self._determine_entity_type(pattern_name, value)
entities.append(Entity(
name=value.strip(),
type=entity_type,
value=value.strip(),
confidence=config.default_entity_confidence
))
except Exception as e:
print(f"Error in pattern {pattern_name}: {e}")
continue
return entities
def _determine_entity_type(self, pattern_name: str, value: str) -> str:
"""Determine entity type from pattern name and value."""
pattern_lower = pattern_name.lower()
if "id" in pattern_lower:
if "challenge" in pattern_lower:
return "challenge_id"
elif "skill" in pattern_lower:
return "skill_id"
elif "member" in pattern_lower or "user" in pattern_lower:
return "user_id"
else:
return "entity_id"
elif "name" in pattern_lower:
if "challenge" in pattern_lower:
return "challenge_name"
elif "skill" in pattern_lower:
return "skill_name"
else:
return "entity_name"
elif "handle" in pattern_lower:
return "user_handle"
else:
# Try to infer from value pattern
if value.isdigit():
return "entity_id"
elif len(value) <= 30 and "_" in value:
return "user_handle"
else:
return "entity_name" |