Spaces:
Paused
Paused
File size: 14,375 Bytes
45ee481 | 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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 | """
Factual Accuracy Module
Verify that generated responses align with CEO's documented positions.
Cross-references claims against source blog content.
Example usage:
checker = FactualAccuracyChecker.from_blogs("data/processed/posts.json")
result = checker.check_response("Generated response with claims...")
"""
import json
import re
from dataclasses import dataclass, field
from pathlib import Path
from typing import Optional
from loguru import logger
try:
from sentence_transformers import SentenceTransformer, util
import numpy as np
SENTENCE_TRANSFORMERS_AVAILABLE = True
except ImportError:
SENTENCE_TRANSFORMERS_AVAILABLE = False
logger.warning("sentence-transformers not available")
@dataclass
class FactualCheckResult:
"""Results from factual accuracy check."""
accuracy_score: float # 0-1 score
verified_claims: list = field(default_factory=list)
unverified_claims: list = field(default_factory=list)
potential_hallucinations: list = field(default_factory=list)
source_citations: list = field(default_factory=list)
def to_dict(self) -> dict:
"""Convert to dictionary."""
return {
"accuracy_score": round(self.accuracy_score, 4),
"num_verified": len(self.verified_claims),
"num_unverified": len(self.unverified_claims),
"num_potential_hallucinations": len(self.potential_hallucinations),
"verified_claims": self.verified_claims[:10],
"unverified_claims": self.unverified_claims[:10],
"potential_hallucinations": self.potential_hallucinations[:5],
}
def passes_threshold(self, threshold: float = 0.95) -> bool:
"""Check if accuracy meets threshold."""
return self.accuracy_score >= threshold
class FactualAccuracyChecker:
"""
Check factual accuracy of generated responses.
Compares claims in generated text against source blog content
to identify potential hallucinations or misrepresentations.
Example:
>>> checker = FactualAccuracyChecker.from_blogs("posts.json")
>>> result = checker.check_response("Response with claims...")
>>> print(f"Accuracy: {result.accuracy_score}")
"""
def __init__(
self,
source_texts: list[dict],
embedding_model: Optional[str] = "all-MiniLM-L6-v2",
similarity_threshold: float = 0.7,
):
"""
Initialize the checker.
Args:
source_texts: List of dicts with 'title' and 'content'
embedding_model: Sentence transformer model
similarity_threshold: Threshold for considering a claim verified
"""
self.source_texts = source_texts
self.similarity_threshold = similarity_threshold
# Build source corpus
self.source_corpus = []
for doc in source_texts:
content = doc.get("content", "")
# Split into paragraphs for finer-grained matching
paragraphs = [p.strip() for p in content.split("\n\n") if p.strip()]
self.source_corpus.extend(paragraphs)
# Load embedding model and encode corpus
self.embedding_model = None
self.corpus_embeddings = None
if SENTENCE_TRANSFORMERS_AVAILABLE and embedding_model:
try:
self.embedding_model = SentenceTransformer(embedding_model)
logger.info(f"Encoding {len(self.source_corpus)} source paragraphs...")
self.corpus_embeddings = self.embedding_model.encode(
self.source_corpus,
convert_to_tensor=True,
show_progress_bar=False,
)
logger.info("Source corpus encoded")
except Exception as e:
logger.warning(f"Failed to load embedding model: {e}")
@classmethod
def from_blogs(
cls,
posts_path: str | Path,
embedding_model: str = "all-MiniLM-L6-v2",
similarity_threshold: float = 0.7,
) -> "FactualAccuracyChecker":
"""
Create checker from parsed blog posts.
Args:
posts_path: Path to posts.json
embedding_model: Sentence transformer model
similarity_threshold: Verification threshold
Returns:
FactualAccuracyChecker instance
"""
with open(posts_path, "r", encoding="utf-8") as f:
posts = json.load(f)
return cls(posts, embedding_model, similarity_threshold)
def check_response(
self,
response: str,
extract_claims: bool = True,
) -> FactualCheckResult:
"""
Check factual accuracy of a generated response.
Args:
response: Generated response text
extract_claims: Whether to extract individual claims
Returns:
FactualCheckResult with accuracy metrics
"""
if extract_claims:
claims = self._extract_claims(response)
else:
# Treat each sentence as a claim
claims = self._split_sentences(response)
if not claims:
return FactualCheckResult(accuracy_score=1.0)
verified = []
unverified = []
hallucinations = []
citations = []
for claim in claims:
is_verified, similarity, source = self._verify_claim(claim)
if is_verified:
verified.append({
"claim": claim,
"similarity": similarity,
"source_excerpt": source[:200] if source else None,
})
if source:
citations.append(source[:100])
else:
# Check if it's a potential hallucination
if self._is_factual_claim(claim):
if similarity < 0.3:
hallucinations.append({
"claim": claim,
"similarity": similarity,
"reason": "No similar content in source",
})
else:
unverified.append({
"claim": claim,
"similarity": similarity,
})
else:
# Opinion or subjective statement - not hallucination
verified.append({
"claim": claim,
"similarity": similarity,
"type": "opinion",
})
# Calculate accuracy
total_factual = len(verified) + len(unverified) + len(hallucinations)
accuracy = len(verified) / total_factual if total_factual > 0 else 1.0
return FactualCheckResult(
accuracy_score=accuracy,
verified_claims=verified,
unverified_claims=unverified,
potential_hallucinations=hallucinations,
source_citations=list(set(citations)),
)
def check_batch(
self,
responses: list[str],
) -> dict:
"""
Check factual accuracy of multiple responses.
Args:
responses: List of generated responses
Returns:
Aggregate metrics
"""
results = [self.check_response(r) for r in responses]
def avg(values):
return sum(values) / len(values) if values else 0
return {
"num_responses": len(results),
"avg_accuracy": avg([r.accuracy_score for r in results]),
"total_verified": sum(len(r.verified_claims) for r in results),
"total_unverified": sum(len(r.unverified_claims) for r in results),
"total_hallucinations": sum(len(r.potential_hallucinations) for r in results),
"pass_rate_0.95": sum(1 for r in results if r.passes_threshold(0.95)) / len(results),
}
def _extract_claims(self, text: str) -> list[str]:
"""Extract factual claims from text."""
sentences = self._split_sentences(text)
claims = []
for sentence in sentences:
# Skip very short sentences
if len(sentence) < 20:
continue
# Skip questions
if sentence.strip().endswith("?"):
continue
# Look for factual indicators
factual_indicators = [
r"\b(is|are|was|were|has|have|had)\b",
r"\b(always|never|every|all|no)\b",
r"\b(percent|%|million|billion)\b",
r"\b(study|research|data|evidence)\b",
r"\b(founded|created|built|developed)\b",
]
is_likely_factual = any(
re.search(pattern, sentence, re.IGNORECASE)
for pattern in factual_indicators
)
if is_likely_factual:
claims.append(sentence)
else:
# Still include as a claim for verification
claims.append(sentence)
return claims
def _split_sentences(self, text: str) -> list[str]:
"""Split text into sentences."""
# Simple sentence splitting
sentences = re.split(r"[.!?]+", text)
return [s.strip() for s in sentences if s.strip()]
def _verify_claim(self, claim: str) -> tuple[bool, float, Optional[str]]:
"""
Verify a claim against source corpus.
Returns:
Tuple of (is_verified, similarity_score, matching_source)
"""
if not self.embedding_model or self.corpus_embeddings is None:
# Fallback to simple text matching
return self._verify_claim_text_match(claim)
try:
claim_embedding = self.embedding_model.encode(
claim, convert_to_tensor=True
)
# Calculate similarities
similarities = util.cos_sim(claim_embedding, self.corpus_embeddings)[0]
max_sim_idx = similarities.argmax().item()
max_similarity = similarities[max_sim_idx].item()
source = self.source_corpus[max_sim_idx] if max_sim_idx < len(self.source_corpus) else None
is_verified = max_similarity >= self.similarity_threshold
return is_verified, max_similarity, source
except Exception as e:
logger.warning(f"Embedding verification failed: {e}")
return self._verify_claim_text_match(claim)
def _verify_claim_text_match(self, claim: str) -> tuple[bool, float, Optional[str]]:
"""Fallback verification using text matching."""
claim_lower = claim.lower()
claim_words = set(re.findall(r"\b\w+\b", claim_lower))
best_match = 0.0
best_source = None
for source in self.source_corpus:
source_lower = source.lower()
source_words = set(re.findall(r"\b\w+\b", source_lower))
# Jaccard similarity
if claim_words and source_words:
intersection = len(claim_words & source_words)
union = len(claim_words | source_words)
similarity = intersection / union
if similarity > best_match:
best_match = similarity
best_source = source
is_verified = best_match >= self.similarity_threshold
return is_verified, best_match, best_source
def _is_factual_claim(self, claim: str) -> bool:
"""Determine if a claim is factual (vs opinion/subjective)."""
opinion_indicators = [
r"\b(i think|i believe|in my opinion|i feel)\b",
r"\b(should|could|might|may)\b",
r"\b(important|interesting|exciting|concerning)\b",
r"\b(best|worst|better|worse)\b",
]
claim_lower = claim.lower()
for pattern in opinion_indicators:
if re.search(pattern, claim_lower):
return False
return True
def main():
"""CLI entry point for testing factual accuracy."""
import argparse
parser = argparse.ArgumentParser(
description="Check factual accuracy of generated responses",
formatter_class=argparse.RawDescriptionHelpFormatter,
epilog="""
Examples:
python factual_accuracy.py --posts posts.json --response "Text to check..."
python factual_accuracy.py --posts posts.json --responses-file outputs.json
""",
)
parser.add_argument("--posts", required=True, help="Parsed posts JSON path")
parser.add_argument("--response", help="Single response to check")
parser.add_argument("--responses-file", help="JSON file with responses")
parser.add_argument("--threshold", type=float, default=0.7, help="Similarity threshold")
args = parser.parse_args()
# Load checker
print(f"Loading source corpus: {args.posts}")
checker = FactualAccuracyChecker.from_blogs(
args.posts,
similarity_threshold=args.threshold,
)
if args.response:
# Check single response
result = checker.check_response(args.response)
print("\n=== Factual Accuracy Check ===")
print(f"Accuracy score: {result.accuracy_score:.2%}")
print(f"Verified claims: {len(result.verified_claims)}")
print(f"Unverified claims: {len(result.unverified_claims)}")
print(f"Potential hallucinations: {len(result.potential_hallucinations)}")
if result.potential_hallucinations:
print("\nPotential hallucinations:")
for h in result.potential_hallucinations[:3]:
print(f" - {h['claim'][:100]}...")
print(f"\nPasses 95% threshold: {result.passes_threshold()}")
elif args.responses_file:
# Check batch
with open(args.responses_file, "r") as f:
data = json.load(f)
responses = [d["response"] if isinstance(d, dict) else d for d in data]
results = checker.check_batch(responses)
print("\n=== Batch Accuracy Check ===")
for key, value in results.items():
print(f"{key}: {value:.4f}" if isinstance(value, float) else f"{key}: {value}")
else:
print("Provide --response or --responses-file")
return 1
return 0
if __name__ == "__main__":
exit(main())
|