Spaces:
Sleeping
Sleeping
File size: 14,107 Bytes
135f0d6 159faf0 135f0d6 159faf0 135f0d6 159faf0 135f0d6 159faf0 135f0d6 159faf0 135f0d6 159faf0 135f0d6 159faf0 135f0d6 159faf0 135f0d6 159faf0 |
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 |
"""
Source Attribution - Citation and source tracking system
This module manages citation generation, source ranking, and quote extraction
for RAG responses with proper source attribution.
"""
import logging
import re
from dataclasses import dataclass
from typing import Any, Dict, List, Optional
logger = logging.getLogger(__name__)
@dataclass
class Citation:
"""Structured citation for source attribution."""
document: str
section: Optional[str] = None
confidence: float = 0.0
excerpt: str = ""
page: Optional[int] = None
url: Optional[str] = None
@dataclass
class Quote:
"""Extracted quote from source document."""
text: str
source_document: str
relevance_score: float
context_before: str = ""
context_after: str = ""
section: Optional[str] = None
@dataclass
class RankedSource:
"""Source document with ranking and metadata."""
document: str
relevance_score: float
reliability_score: float
excerpt: str
metadata: Dict[str, Any]
rank: int = 0
class SourceAttributor:
"""
Manages citation generation and source tracking for RAG responses.
Provides:
- Structured citation formatting
- Source ranking by relevance and reliability
- Quote extraction from source documents
- Citation validation and verification
"""
def __init__(self, config: Optional[Dict[str, Any]] = None):
"""
Initialize SourceAttributor with configuration.
Args:
config: Configuration dictionary for attribution settings
"""
self.config = config or self._get_default_config()
logger.info("SourceAttributor initialized")
def _get_default_config(self) -> Dict[str, Any]:
"""Get default attribution configuration."""
return {
"max_citations": 5,
"min_confidence_for_citation": 0.3,
"citation_format": "numbered", # "numbered", "parenthetical", "footnote"
"include_excerpts": True,
"max_excerpt_length": 150,
"require_document_names": True,
"prefer_specific_sections": True,
}
def generate_citations(self, response: str, sources: List[Dict[str, Any]]) -> List[Citation]:
"""
Generate proper citations for response based on sources.
Args:
response: Generated response text
sources: Source documents with metadata
Returns:
List of Citation objects for the response
"""
try:
citations = []
# Rank sources by relevance and reliability
ranked_sources = self.rank_sources(sources, [])
# Generate citations for top sources
for i, ranked_source in enumerate(ranked_sources[: self.config["max_citations"]]):
if ranked_source.relevance_score >= self.config["min_confidence_for_citation"]:
citation = self._create_citation(ranked_source, i + 1)
citations.append(citation)
# Ensure citations are properly embedded in response
self._validate_citation_presence(response, citations)
logger.debug(f"Generated {len(citations)} citations")
return citations
except Exception as e:
logger.error(f"Citation generation error: {e}")
return []
def extract_quotes(self, response: str, documents: List[Dict[str, Any]]) -> List[Quote]:
"""
Extract relevant quotes from source documents.
Args:
response: Generated response text
documents: Source documents to extract quotes from
Returns:
List of Quote objects with extracted text
"""
try:
quotes = []
for doc in documents:
content = doc.get("content", "")
document_name = doc.get("metadata", {}).get("filename", "unknown")
# Find quotes that appear in both response and document
extracted_quotes = self._find_matching_quotes(response, content)
for quote_text in extracted_quotes:
relevance = self._calculate_quote_relevance(quote_text, response)
quote = Quote(
text=quote_text,
source_document=document_name,
relevance_score=relevance,
section=doc.get("metadata", {}).get("section"),
)
quotes.append(quote)
# Sort by relevance
quotes.sort(key=lambda q: q.relevance_score, reverse=True)
logger.debug(f"Extracted {len(quotes)} quotes")
return quotes
except Exception as e:
logger.error(f"Quote extraction error: {e}")
return []
def rank_sources(self, sources: List[Dict[str, Any]], relevance_scores: List[float]) -> List[RankedSource]:
"""
Rank sources by relevance and reliability.
Args:
sources: Source documents with metadata
relevance_scores: Pre-calculated relevance scores (optional)
Returns:
List of RankedSource objects sorted by ranking
"""
try:
ranked_sources = []
for i, source in enumerate(sources):
# Use provided relevance or calculate
if i < len(relevance_scores):
relevance = relevance_scores[i]
else:
relevance = source.get("relevance_score", 0.5)
# Calculate reliability score
reliability = self._calculate_reliability(source)
# Create ranked source
ranked_source = RankedSource(
document=source.get("metadata", {}).get("filename", "unknown"),
relevance_score=relevance,
reliability_score=reliability,
excerpt=self._create_excerpt(source),
metadata=source.get("metadata", {}),
)
ranked_sources.append(ranked_source)
# Sort by combined score (relevance + reliability)
ranked_sources.sort(
key=lambda rs: (rs.relevance_score + rs.reliability_score) / 2,
reverse=True,
)
# Assign ranks
for i, ranked_source in enumerate(ranked_sources):
ranked_source.rank = i + 1
logger.debug(f"Ranked {len(ranked_sources)} sources")
return ranked_sources
except Exception as e:
logger.error(f"Source ranking error: {e}")
return []
def format_citation_text(self, citations: List[Citation]) -> str:
"""
Format citations as text for inclusion in response.
Args:
citations: List of Citation objects
Returns:
Formatted citation text
"""
if not citations:
return ""
citation_format = self.config["citation_format"]
if citation_format == "numbered":
return self._format_numbered_citations(citations)
elif citation_format == "parenthetical":
return self._format_parenthetical_citations(citations)
elif citation_format == "footnote":
return self._format_footnote_citations(citations)
else:
return self._format_numbered_citations(citations)
def validate_citations(self, response: str, citations: List[Citation]) -> Dict[str, bool]:
"""
Validate that citations are properly referenced in response.
Args:
response: Response text to validate
citations: Citations that should be referenced
Returns:
Dictionary mapping citation to validation status
"""
validation_results = {}
for citation in citations:
is_valid = self._is_citation_referenced(response, citation)
validation_results[citation.document] = is_valid
return validation_results
def _create_citation(self, ranked_source: RankedSource, number: int) -> Citation:
"""Create Citation object from ranked source."""
return Citation(
document=ranked_source.document,
section=ranked_source.metadata.get("section"),
confidence=ranked_source.relevance_score,
excerpt=ranked_source.excerpt,
page=ranked_source.metadata.get("page"),
url=ranked_source.metadata.get("url"),
)
def _calculate_reliability(self, source: Dict[str, Any]) -> float:
"""Calculate reliability score for source document."""
# Base reliability
reliability = 0.7
# Boost for official documents
filename = source.get("metadata", {}).get("filename", "").lower()
if any(term in filename for term in ["policy", "handbook", "guideline", "procedure", "manual"]):
reliability += 0.2
# Boost for recent documents (if timestamp available)
# This would need timestamp metadata
# if 'last_modified' in source.get('metadata', {}):
# # Add recency bonus
# pass
# Boost for documents with clear structure
content = source.get("content", "")
if any(marker in content.lower() for marker in ["section", "article", "paragraph", "clause"]):
reliability += 0.1
return min(reliability, 1.0)
def _create_excerpt(self, source: Dict[str, Any]) -> str:
"""Create excerpt from source document."""
content = source.get("content", "")
max_length = self.config["max_excerpt_length"]
if len(content) <= max_length:
return content
# Try to find a good breaking point
excerpt = content[:max_length]
last_sentence = excerpt.rfind(".")
last_space = excerpt.rfind(" ")
if last_sentence > max_length * 0.7:
return excerpt[: last_sentence + 1]
elif last_space > max_length * 0.8:
return excerpt[:last_space] + "..."
else:
return excerpt + "..."
def _find_matching_quotes(self, response: str, document_content: str) -> List[str]:
"""Find quotes that appear in both response and document."""
quotes = []
# Look for phrases that appear in both
response_sentences = [s.strip() for s in response.split(".") if s.strip()]
doc_sentences = [s.strip() for s in document_content.split(".") if s.strip()]
for resp_sent in response_sentences:
for doc_sent in doc_sentences:
# Check for substantial overlap
if len(resp_sent) > 20 and len(doc_sent) > 20:
if self._calculate_sentence_similarity(resp_sent, doc_sent) > 0.7:
quotes.append(doc_sent)
return list(set(quotes)) # Remove duplicates
def _calculate_sentence_similarity(self, sent1: str, sent2: str) -> float:
"""Calculate similarity between two sentences."""
words1 = set(sent1.lower().split())
words2 = set(sent2.lower().split())
intersection = words1.intersection(words2)
union = words1.union(words2)
if not union:
return 0.0
return len(intersection) / len(union)
def _calculate_quote_relevance(self, quote: str, response: str) -> float:
"""Calculate relevance of quote to response."""
return self._calculate_sentence_similarity(quote, response)
def _validate_citation_presence(self, response: str, citations: List[Citation]) -> None:
"""Validate that citations are present in response."""
if not self.config["require_document_names"]:
return
for citation in citations:
if citation.document.lower() not in response.lower():
logger.warning(f"Citation {citation.document} not found in response")
def _format_numbered_citations(self, citations: List[Citation]) -> str:
"""Format citations in numbered format."""
if not citations:
return ""
formatted = "\n\n**Sources:**\n"
for i, citation in enumerate(citations, 1):
formatted += f"{i}. {citation.document}"
if citation.section:
formatted += f" ({citation.section})"
if self.config["include_excerpts"] and citation.excerpt:
formatted += f'\n "{citation.excerpt}"'
formatted += "\n"
return formatted
def _format_parenthetical_citations(self, citations: List[Citation]) -> str:
"""Format citations in parenthetical format."""
if not citations:
return ""
# Simple format: (Document1, Document2)
doc_names = [citation.document for citation in citations]
return f" ({', '.join(doc_names)})"
def _format_footnote_citations(self, citations: List[Citation]) -> str:
"""Format citations as footnotes."""
if not citations:
return ""
formatted = "\n\n**References:**\n"
for i, citation in enumerate(citations, 1):
formatted += f"[{i}] {citation.document}"
if citation.section:
formatted += f", {citation.section}"
formatted += "\n"
return formatted
def _is_citation_referenced(self, response: str, citation: Citation) -> bool:
"""Check if citation is properly referenced in response."""
response_lower = response.lower()
doc_name_lower = citation.document.lower()
# Look for document name mentions
if doc_name_lower in response_lower:
return True
# Look for citation patterns
citation_patterns = [
rf"\[.*{re.escape(citation.document)}.*\]",
rf"\(.*{re.escape(citation.document)}.*\)",
]
return any(re.search(pattern, response, re.IGNORECASE) for pattern in citation_patterns)
|