Spaces:
Running
Running
File size: 13,528 Bytes
5301c48 |
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 |
"""Data Deduplicator for Dataset Generation
This module provides advanced deduplication capabilities for generated datasets
using semantic embeddings to detect and remove similar content.
"""
from typing import List, Dict, Any, Optional, Union, Tuple, Callable
from .embedding_manager import EmbeddingManager
from .similarity_checker import SimilarityChecker
from starfish.common.logger import get_logger
import json
import hashlib
logger = get_logger(__name__)
class DataDeduplicator:
"""
Advanced deduplication for generated datasets using semantic embeddings.
Features:
- Semantic deduplication using embeddings
- Exact match deduplication using hashing
- Field-specific deduplication strategies
- Preservation of highest quality items
- Detailed deduplication reports
"""
def __init__(
self,
embedding_manager: Optional[EmbeddingManager] = None,
similarity_threshold: float = 0.9,
exact_match_fields: Optional[List[str]] = None,
semantic_fields: Optional[List[str]] = None,
quality_scorer: Optional[Callable] = None,
):
"""
Initialize the DataDeduplicator.
Args:
embedding_manager: Pre-configured EmbeddingManager instance
similarity_threshold: Threshold for semantic similarity (0-1)
exact_match_fields: Fields to check for exact matches
semantic_fields: Fields to check for semantic similarity
quality_scorer: Function to score item quality for keeping best duplicates
"""
self.embedding_manager = embedding_manager or EmbeddingManager()
self.similarity_checker = SimilarityChecker(embedding_manager=self.embedding_manager, similarity_threshold=similarity_threshold)
self.similarity_threshold = similarity_threshold
self.exact_match_fields = exact_match_fields or ["id", "uuid"]
self.semantic_fields = semantic_fields or ["text", "query", "question", "content", "prompt", "response", "answer"]
self.quality_scorer = quality_scorer or self._default_quality_scorer
logger.info(f"DataDeduplicator initialized with threshold={similarity_threshold}")
def _default_quality_scorer(self, item: Dict[str, Any]) -> float:
"""
Default quality scoring function.
Args:
item: Data item to score
Returns:
Quality score (higher is better)
"""
score = 0.0
# Length bonus for longer content
for field in self.semantic_fields:
if field in item and isinstance(item[field], str):
score += len(item[field]) * 0.001
# Completeness bonus
non_empty_fields = sum(1 for v in item.values() if v is not None and str(v).strip())
score += non_empty_fields * 0.1
# Specific quality indicators
if "score" in item:
score += float(item.get("score", 0))
if "confidence" in item:
score += float(item.get("confidence", 0))
return score
def _extract_exact_match_signature(self, item: Dict[str, Any]) -> str:
"""
Create a signature for exact match detection.
Args:
item: Data item
Returns:
Hash signature for exact matching
"""
signature_parts = []
for field in self.exact_match_fields:
if field in item and item[field] is not None:
signature_parts.append(f"{field}:{item[field]}")
# Fallback to content hash if no exact match fields
if not signature_parts:
content = json.dumps(item, sort_keys=True, ensure_ascii=False)
return hashlib.md5(content.encode()).hexdigest()
signature = "|".join(signature_parts)
return hashlib.md5(signature.encode()).hexdigest()
def _extract_semantic_content(self, item: Dict[str, Any]) -> str:
"""
Extract semantic content for similarity comparison.
Args:
item: Data item
Returns:
Combined semantic content
"""
return self.similarity_checker.extract_text(item)
def deduplicate_exact(self, items: List[Dict[str, Any]], keep_best: bool = True) -> Tuple[List[Dict[str, Any]], Dict[str, Any]]:
"""
Remove exact duplicates based on specified fields.
Args:
items: List of data items
keep_best: Whether to keep the highest quality item in each duplicate group
Returns:
Tuple of (deduplicated_items, report)
"""
if not items:
return [], {"exact_duplicates_removed": 0, "groups": []}
# Group items by signature
signature_groups = {}
for i, item in enumerate(items):
signature = self._extract_exact_match_signature(item)
if signature not in signature_groups:
signature_groups[signature] = []
signature_groups[signature].append((i, item))
# Process groups
deduplicated_items = []
duplicate_groups = []
total_removed = 0
for signature, group in signature_groups.items():
if len(group) == 1:
# No duplicates
deduplicated_items.append(group[0][1])
else:
# Duplicates found
duplicate_groups.append([idx for idx, _ in group])
total_removed += len(group) - 1
if keep_best:
# Keep the highest quality item
best_item = max(group, key=lambda x: self.quality_scorer(x[1]))[1]
deduplicated_items.append(best_item)
else:
# Keep the first item
deduplicated_items.append(group[0][1])
report = {"exact_duplicates_removed": total_removed, "groups": duplicate_groups, "original_count": len(items), "final_count": len(deduplicated_items)}
logger.info(f"Exact deduplication: {len(items)} -> {len(deduplicated_items)} ({total_removed} removed)")
return deduplicated_items, report
def deduplicate_semantic(
self, items: List[Dict[str, Any]], threshold: Optional[float] = None, keep_best: bool = True
) -> Tuple[List[Dict[str, Any]], Dict[str, Any]]:
"""
Remove semantic duplicates using embedding similarity.
Args:
items: List of data items
threshold: Custom similarity threshold
keep_best: Whether to keep the highest quality item in each duplicate group
Returns:
Tuple of (deduplicated_items, report)
"""
if not items:
return [], {"semantic_duplicates_removed": 0, "groups": []}
threshold = threshold or self.similarity_threshold
# Extract semantic content
semantic_contents = [self._extract_semantic_content(item) for item in items]
# Find duplicate groups using embedding similarity
duplicate_groups = self.embedding_manager.find_duplicates(semantic_contents, threshold)
# Track which items to keep
items_to_remove = set()
processed_groups = []
for group in duplicate_groups:
if len(group) > 1:
processed_groups.append(group)
if keep_best:
# Find the best item in the group
group_items = [(idx, items[idx]) for idx in group]
best_idx = max(group_items, key=lambda x: self.quality_scorer(x[1]))[0]
# Remove all except the best
for idx in group:
if idx != best_idx:
items_to_remove.add(idx)
else:
# Remove all except the first
for idx in group[1:]:
items_to_remove.add(idx)
# Create deduplicated list
deduplicated_items = [item for i, item in enumerate(items) if i not in items_to_remove]
report = {
"semantic_duplicates_removed": len(items_to_remove),
"groups": processed_groups,
"similarity_threshold": threshold,
"original_count": len(items),
"final_count": len(deduplicated_items),
}
logger.info(f"Semantic deduplication: {len(items)} -> {len(deduplicated_items)} ({len(items_to_remove)} removed)")
return deduplicated_items, report
def deduplicate_comprehensive(
self, items: List[Dict[str, Any]], exact_first: bool = True, semantic_threshold: Optional[float] = None, keep_best: bool = True
) -> Tuple[List[Dict[str, Any]], Dict[str, Any]]:
"""
Perform comprehensive deduplication using both exact and semantic methods.
Args:
items: List of data items
exact_first: Whether to perform exact deduplication first
semantic_threshold: Custom similarity threshold for semantic deduplication
keep_best: Whether to keep the highest quality item in each duplicate group
Returns:
Tuple of (deduplicated_items, comprehensive_report)
"""
if not items:
return [], {"total_removed": 0, "stages": []}
current_items = items.copy()
reports = []
if exact_first:
# Stage 1: Exact deduplication
current_items, exact_report = self.deduplicate_exact(current_items, keep_best)
exact_report["stage"] = "exact"
reports.append(exact_report)
# Stage 2: Semantic deduplication
current_items, semantic_report = self.deduplicate_semantic(current_items, semantic_threshold, keep_best)
semantic_report["stage"] = "semantic"
reports.append(semantic_report)
else:
# Stage 1: Semantic deduplication
current_items, semantic_report = self.deduplicate_semantic(current_items, semantic_threshold, keep_best)
semantic_report["stage"] = "semantic"
reports.append(semantic_report)
# Stage 2: Exact deduplication
current_items, exact_report = self.deduplicate_exact(current_items, keep_best)
exact_report["stage"] = "exact"
reports.append(exact_report)
total_removed = len(items) - len(current_items)
comprehensive_report = {
"total_removed": total_removed,
"original_count": len(items),
"final_count": len(current_items),
"reduction_percentage": (total_removed / len(items)) * 100 if items else 0,
"stages": reports,
"processing_order": ["exact", "semantic"] if exact_first else ["semantic", "exact"],
}
logger.info(
f"Comprehensive deduplication: {len(items)} -> {len(current_items)} "
f"({total_removed} removed, {comprehensive_report['reduction_percentage']:.1f}% reduction)"
)
return current_items, comprehensive_report
def analyze_duplicates(self, items: List[Dict[str, Any]], semantic_threshold: Optional[float] = None) -> Dict[str, Any]:
"""
Analyze duplicate patterns without removing items.
Args:
items: List of data items to analyze
semantic_threshold: Custom similarity threshold
Returns:
Analysis report with duplicate statistics
"""
if not items:
return {"total_items": 0, "analysis": "No items to analyze"}
# Exact duplicate analysis
signature_counts = {}
for item in items:
signature = self._extract_exact_match_signature(item)
signature_counts[signature] = signature_counts.get(signature, 0) + 1
exact_duplicates = sum(count - 1 for count in signature_counts.values() if count > 1)
exact_groups = sum(1 for count in signature_counts.values() if count > 1)
# Semantic duplicate analysis
semantic_contents = [self._extract_semantic_content(item) for item in items]
threshold = semantic_threshold or self.similarity_threshold
duplicate_groups = self.embedding_manager.find_duplicates(semantic_contents, threshold)
semantic_duplicates = sum(len(group) - 1 for group in duplicate_groups if len(group) > 1)
semantic_groups = len([group for group in duplicate_groups if len(group) > 1])
# Diversity analysis
diversity_metrics = self.similarity_checker.check_diversity_batch(items)
analysis = {
"total_items": len(items),
"exact_duplicates": {"count": exact_duplicates, "groups": exact_groups, "percentage": (exact_duplicates / len(items)) * 100 if items else 0},
"semantic_duplicates": {
"count": semantic_duplicates,
"groups": semantic_groups,
"percentage": (semantic_duplicates / len(items)) * 100 if items else 0,
"threshold": threshold,
},
"diversity_metrics": diversity_metrics,
"quality_scores": {
"min": min(self.quality_scorer(item) for item in items),
"max": max(self.quality_scorer(item) for item in items),
"avg": sum(self.quality_scorer(item) for item in items) / len(items),
},
}
logger.info(f"Duplicate analysis: {exact_duplicates} exact, {semantic_duplicates} semantic duplicates found")
return analysis
|