File size: 13,773 Bytes
1cc481e | 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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 | """
Dialog Manager — Conversation State Tracking & Context Management
=================================================================
Manages multi-turn dialog state for Bengali public service conversations.
Responsibilities:
- Track conversation history (user + agent turns)
- Maintain slot/entity memory across turns
- Determine dialog acts (greet → inform → query → confirm → close)
- Build context windows for the response generation model
- Handle domain switching and topic transitions
Designed to work with:
- JointIntentNER model (NLU component)
- BanglaT5 response generation (NLG component)
"""
import json
from typing import Dict, List, Optional, Tuple
from dataclasses import dataclass, field, asdict
from enum import Enum
from datetime import datetime
# ============================================================================
# DIALOG STATES
# ============================================================================
class DialogState(Enum):
"""High-level dialog states."""
IDLE = "idle" # No active conversation
GREETING = "greeting" # Initial greeting phase
INFORMATION = "information" # Providing/collecting information
QUERY = "query" # User asking questions
CONFIRMATION = "confirmation" # Confirming details
CLOSING = "closing" # Wrapping up conversation
ESCALATION = "escalation" # Needs human agent
# Intent-to-state mapping
INTENT_STATE_MAP = {
"greeting": DialogState.GREETING,
"farewell": DialogState.CLOSING,
"thanks": DialogState.CLOSING,
"passport_application": DialogState.INFORMATION,
"passport_renewal": DialogState.INFORMATION,
"passport_status": DialogState.QUERY,
"passport_fee": DialogState.QUERY,
"nid_application": DialogState.INFORMATION,
"nid_correction": DialogState.INFORMATION,
"nid_status": DialogState.QUERY,
"utility_bill_payment": DialogState.INFORMATION,
"utility_new_connection": DialogState.INFORMATION,
"utility_complaint": DialogState.QUERY,
"welfare_application": DialogState.INFORMATION,
"welfare_eligibility": DialogState.QUERY,
"welfare_status": DialogState.QUERY,
"general_inquiry": DialogState.QUERY,
"complaint": DialogState.ESCALATION,
}
# ============================================================================
# SLOT DEFINITIONS PER DOMAIN
# ============================================================================
DOMAIN_SLOTS = {
"passport": [
"applicant_name", "nid_number", "date_of_birth",
"passport_type", "application_type", "fee_amount",
],
"nid": [
"applicant_name", "date_of_birth", "voter_area",
"correction_field", "nid_number",
],
"utilities": [
"account_number", "bill_type", "payment_method",
"complaint_type", "connection_type", "area",
],
"welfare": [
"applicant_name", "age", "scheme_name",
"eligibility_status", "application_id",
],
"general": [
"topic", "query_type",
],
}
# ============================================================================
# DATA STRUCTURES
# ============================================================================
@dataclass
class Turn:
"""A single turn in the conversation."""
role: str # "citizen" or "agent"
text: str # The utterance text
intent: Optional[str] = None
entities: Optional[Dict[str, str]] = None
timestamp: Optional[str] = None
def to_dict(self) -> Dict:
return asdict(self)
@dataclass
class ConversationState:
"""Full state of a conversation."""
dialog_id: str
domain: str = "general"
state: DialogState = DialogState.IDLE
turns: List[Turn] = field(default_factory=list)
slots: Dict[str, Optional[str]] = field(default_factory=dict)
turn_count: int = 0
confidence_scores: List[float] = field(default_factory=list)
created_at: str = field(default_factory=lambda: datetime.now().isoformat())
def to_dict(self) -> Dict:
d = asdict(self)
d["state"] = self.state.value
return d
# ============================================================================
# DIALOG MANAGER
# ============================================================================
class DialogManager:
"""
Manages dialog state for multi-turn Bengali public service conversations.
The dialog manager sits between NLU (intent + entities) and NLG (response
generation), maintaining conversation context and determining the system's
next action.
Architecture:
User Input → NLU → DialogManager.update() → context → NLG → Response
"""
def __init__(self, max_context_turns: int = 5, max_turns: int = 20):
"""
Args:
max_context_turns: Number of recent turns to include in context
window for response generation.
max_turns: Maximum turns before suggesting escalation.
"""
self.max_context_turns = max_context_turns
self.max_turns = max_turns
self.conversations: Dict[str, ConversationState] = {}
def start_conversation(self, dialog_id: str, domain: str = "general") -> ConversationState:
"""Initialize a new conversation."""
conv = ConversationState(
dialog_id=dialog_id,
domain=domain,
state=DialogState.IDLE,
slots={slot: None for slot in DOMAIN_SLOTS.get(domain, [])},
)
self.conversations[dialog_id] = conv
return conv
def get_conversation(self, dialog_id: str) -> Optional[ConversationState]:
"""Retrieve an existing conversation."""
return self.conversations.get(dialog_id)
def update(
self,
dialog_id: str,
user_text: str,
intent: str,
entities: Dict[str, str],
confidence: float = 1.0,
) -> Tuple[ConversationState, str]:
"""
Process a user turn and update dialog state.
Args:
dialog_id: Conversation identifier
user_text: The user's utterance
intent: Predicted intent from NLU
entities: Extracted entities from NLU {entity_type: value}
confidence: Intent classification confidence score
Returns:
(updated_state, context_for_nlg)
"""
conv = self.conversations.get(dialog_id)
if conv is None:
conv = self.start_conversation(dialog_id)
# 1. Record the user turn
user_turn = Turn(
role="citizen",
text=user_text,
intent=intent,
entities=entities if entities else None,
timestamp=datetime.now().isoformat(),
)
conv.turns.append(user_turn)
conv.turn_count += 1
conv.confidence_scores.append(confidence)
# 2. Update domain based on intent (if domain-specific)
new_domain = self._infer_domain(intent)
if new_domain and new_domain != conv.domain:
conv.domain = new_domain
# Re-initialize slots for new domain
conv.slots = {slot: None for slot in DOMAIN_SLOTS.get(new_domain, [])}
# 3. Update dialog state
conv.state = self._transition_state(conv, intent, confidence)
# 4. Fill slots from entities
self._fill_slots(conv, entities)
# 5. Build context for response generation
context = self._build_context(conv)
return conv, context
def add_agent_response(self, dialog_id: str, response_text: str):
"""Record the agent's response in conversation history."""
conv = self.conversations.get(dialog_id)
if conv is None:
return
agent_turn = Turn(
role="agent",
text=response_text,
timestamp=datetime.now().isoformat(),
)
conv.turns.append(agent_turn)
def get_filled_slots(self, dialog_id: str) -> Dict[str, str]:
"""Return slots that have been filled."""
conv = self.conversations.get(dialog_id)
if conv is None:
return {}
return {k: v for k, v in conv.slots.items() if v is not None}
def get_missing_slots(self, dialog_id: str) -> List[str]:
"""Return slots that still need to be filled."""
conv = self.conversations.get(dialog_id)
if conv is None:
return []
return [k for k, v in conv.slots.items() if v is None]
def should_escalate(self, dialog_id: str) -> bool:
"""Check if conversation should be escalated to human agent."""
conv = self.conversations.get(dialog_id)
if conv is None:
return False
# Escalate if: explicit complaint, too many turns, or low confidence
if conv.state == DialogState.ESCALATION:
return True
if conv.turn_count > self.max_turns:
return True
if len(conv.confidence_scores) >= 3:
recent = conv.confidence_scores[-3:]
if all(c < 0.5 for c in recent):
return True
return False
def end_conversation(self, dialog_id: str) -> Optional[Dict]:
"""End a conversation and return its summary."""
conv = self.conversations.pop(dialog_id, None)
if conv is None:
return None
return {
"dialog_id": dialog_id,
"domain": conv.domain,
"total_turns": conv.turn_count,
"final_state": conv.state.value,
"filled_slots": self.get_filled_slots(dialog_id),
"avg_confidence": (
sum(conv.confidence_scores) / len(conv.confidence_scores)
if conv.confidence_scores else 0
),
}
# ------------------------------------------------------------------
# Internal Methods
# ------------------------------------------------------------------
def _infer_domain(self, intent: str) -> Optional[str]:
"""Infer domain from intent name."""
if intent.startswith("passport"):
return "passport"
elif intent.startswith("nid"):
return "nid"
elif intent.startswith("utility"):
return "utilities"
elif intent.startswith("welfare"):
return "welfare"
return None
def _transition_state(
self, conv: ConversationState, intent: str, confidence: float
) -> DialogState:
"""Determine next dialog state based on current state + intent."""
# Low confidence → stay in current state (don't make wrong transitions)
if confidence < 0.3:
return conv.state
# Map intent to target state
target = INTENT_STATE_MAP.get(intent, DialogState.QUERY)
# State transition rules
current = conv.state
if current == DialogState.IDLE:
return target
if current == DialogState.GREETING:
# After greeting, move to whatever the user wants
if target in (DialogState.GREETING, DialogState.CLOSING):
return target
return target
if current == DialogState.CLOSING:
# If user continues after farewell, re-open
if target not in (DialogState.CLOSING,):
return target
return DialogState.CLOSING
# Default: follow the intent mapping
return target
def _fill_slots(self, conv: ConversationState, entities: Dict[str, str]):
"""Fill conversation slots from extracted entities."""
if not entities:
return
# Map NER entity types to slot names
entity_slot_map = {
"PERSON": "applicant_name",
"NID": "nid_number",
"DATE": "date_of_birth",
"MONEY": "fee_amount",
"LOCATION": "area",
"ACCOUNT": "account_number",
"AGE": "age",
"SCHEME": "scheme_name",
"DOCUMENT": "passport_type",
}
for entity_type, value in entities.items():
slot_name = entity_slot_map.get(entity_type)
if slot_name and slot_name in conv.slots:
conv.slots[slot_name] = value
def _build_context(self, conv: ConversationState) -> str:
"""
Build context string for response generation model.
Takes the last N turns and formats them as the model expects.
"""
# Get recent turns (up to max_context_turns)
recent_turns = conv.turns[-self.max_context_turns:]
# Format as "role: text" pairs
context_parts = []
for turn in recent_turns:
if turn.role == "citizen":
context_parts.append(f"নাগরিক: {turn.text}")
else:
context_parts.append(f"এজেন্ট: {turn.text}")
return " ".join(context_parts)
def get_state_summary(self, dialog_id: str) -> Dict:
"""Get a summary of current conversation state (for debugging/logging)."""
conv = self.conversations.get(dialog_id)
if conv is None:
return {"error": "Conversation not found"}
return {
"dialog_id": dialog_id,
"domain": conv.domain,
"state": conv.state.value,
"turn_count": conv.turn_count,
"filled_slots": {k: v for k, v in conv.slots.items() if v is not None},
"missing_slots": [k for k, v in conv.slots.items() if v is None],
"should_escalate": self.should_escalate(dialog_id),
"last_intent": (
conv.turns[-1].intent if conv.turns and conv.turns[-1].intent else None
),
}
|