Spaces:
Running
Running
File size: 42,970 Bytes
c2ea5ed |
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833 834 835 836 837 838 839 840 841 842 843 844 845 846 847 848 849 850 851 852 853 854 855 856 857 858 859 860 861 862 863 864 865 866 867 868 869 870 871 872 873 874 875 876 877 878 879 880 881 882 883 884 885 886 887 888 889 890 891 892 893 894 895 896 897 898 899 900 901 902 903 904 905 906 907 908 909 910 911 912 913 914 915 916 917 918 919 920 921 922 923 924 925 926 927 928 929 930 931 932 933 934 935 936 937 938 939 940 941 942 943 944 945 946 947 948 949 950 951 952 953 954 955 956 957 958 959 960 961 962 963 964 965 966 |
"""
Knowledge Graph Comparator
This module provides functionality to compare two knowledge graphs from the database
and generate comprehensive comparison metrics including structural similarity,
semantic similarity, and entity/relationship overlap analysis.
"""
import json
import logging
import numpy as np
from typing import Dict, List, Any, Tuple, Set, Optional
from dataclasses import dataclass
import os
import openai
from sklearn.metrics.pairwise import cosine_similarity
from scipy.optimize import linear_sum_assignment
import time
import hashlib
import pickle
# Configure OpenAI for embeddings
openai.api_key = os.environ.get("OPENAI_API_KEY")
@dataclass
class GraphComparisonMetrics:
"""Comprehensive metrics for comparing two knowledge graphs"""
# Entity comparison metrics
entity_overlap_count: int
entity_unique_to_graph1: int
entity_unique_to_graph2: int
entity_overlap_ratio: float
entity_semantic_similarity: float
# Relation comparison metrics
relation_overlap_count: int
relation_unique_to_graph1: int
relation_unique_to_graph2: int
relation_overlap_ratio: float
relation_semantic_similarity: float
# Structural metrics
graph1_density: float
graph2_density: float
density_difference: float
common_patterns_count: int
# Type distribution metrics
entity_type_similarity: float
relation_type_similarity: float
# Overall similarity scores
structural_similarity: float
content_similarity: float
overall_similarity: float
# Additional statistics
graph1_stats: Dict[str, Any]
graph2_stats: Dict[str, Any]
def to_dict(self) -> Dict[str, Any]:
"""Convert metrics to dictionary for JSON serialization"""
return {
"entity_metrics": {
"overlap_count": self.entity_overlap_count,
"unique_to_graph1": self.entity_unique_to_graph1,
"unique_to_graph2": self.entity_unique_to_graph2,
"overlap_ratio": self.entity_overlap_ratio,
"semantic_similarity": self.entity_semantic_similarity
},
"relation_metrics": {
"overlap_count": self.relation_overlap_count,
"unique_to_graph1": self.relation_unique_to_graph1,
"unique_to_graph2": self.relation_unique_to_graph2,
"overlap_ratio": self.relation_overlap_ratio,
"semantic_similarity": self.relation_semantic_similarity
},
"structural_metrics": {
"graph1_density": self.graph1_density,
"graph2_density": self.graph2_density,
"density_difference": self.density_difference,
"common_patterns_count": self.common_patterns_count
},
"type_distribution_metrics": {
"entity_type_similarity": self.entity_type_similarity,
"relation_type_similarity": self.relation_type_similarity
},
"overall_metrics": {
"structural_similarity": self.structural_similarity,
"content_similarity": self.content_similarity,
"overall_similarity": self.overall_similarity
},
"graph_statistics": {
"graph1_stats": self.graph1_stats,
"graph2_stats": self.graph2_stats
}
}
class KnowledgeGraphComparator:
"""Main class for comparing two knowledge graphs"""
def __init__(self, similarity_threshold: float = 0.7, semantic_threshold: float = 0.75, use_cache: bool = True):
"""
Initialize the comparator.
Args:
similarity_threshold: Threshold for semantic similarity matching (0.7 = 70%)
semantic_threshold: Threshold for semantic overlap detection for same-trace graphs (0.75 = 75%)
Higher values = more strict/precise matching
0.9+ = Very high similarity (almost identical)
0.8-0.9 = High similarity (very likely same concept)
0.75-0.8 = Good similarity (probably same with minor variations)
0.65-0.75 = Moderate similarity (related but potentially different)
0.5-0.65 = Low similarity (loosely related)
use_cache: Whether to use embedding cache (default: True)
"""
self.similarity_threshold = similarity_threshold
self.semantic_threshold = semantic_threshold
self.use_cache = use_cache
# Initialize embedding cache
self.embedding_cache = {}
self.cache_file = "cache/embeddings_cache.pkl"
if self.use_cache:
self._load_embedding_cache()
else:
logging.info("Cache disabled for this comparison")
def _load_embedding_cache(self):
"""Load embedding cache from file"""
try:
os.makedirs(os.path.dirname(self.cache_file), exist_ok=True)
if os.path.exists(self.cache_file):
with open(self.cache_file, 'rb') as f:
self.embedding_cache = pickle.load(f)
logging.info(f"Loaded {len(self.embedding_cache)} cached embeddings")
else:
self.embedding_cache = {}
except Exception as e:
logging.error(f"Error loading embedding cache: {e}")
self.embedding_cache = {}
def _save_embedding_cache(self):
"""Save embedding cache to file"""
try:
os.makedirs(os.path.dirname(self.cache_file), exist_ok=True)
with open(self.cache_file, 'wb') as f:
pickle.dump(self.embedding_cache, f)
logging.debug(f"Saved {len(self.embedding_cache)} embeddings to cache")
except Exception as e:
logging.error(f"Error saving embedding cache: {e}")
def _get_text_hash(self, text: str) -> str:
"""Get hash for text to use as cache key"""
return hashlib.md5(text.encode('utf-8')).hexdigest()
def clear_embedding_cache(self):
"""Clear all cached embeddings"""
try:
self.embedding_cache = {}
if os.path.exists(self.cache_file):
os.remove(self.cache_file)
logging.info("Embedding cache cleared successfully")
return True
except Exception as e:
logging.error(f"Error clearing embedding cache: {e}")
return False
def get_cache_info(self) -> Dict[str, Any]:
"""Get information about the current cache"""
cache_size = len(self.embedding_cache)
file_exists = os.path.exists(self.cache_file)
file_size = 0
if file_exists:
try:
file_size = os.path.getsize(self.cache_file)
except Exception:
file_size = 0
return {
"cache_entries": cache_size,
"cache_file_exists": file_exists,
"cache_file_size_bytes": file_size,
"cache_file_size_mb": round(file_size / (1024 * 1024), 2) if file_size > 0 else 0
}
def get_embedding(self, text: str) -> np.ndarray:
"""Get embedding for text using OpenAI text-embedding-3-small with optional caching"""
if not text or not text.strip():
return np.zeros(1536)
# Check cache first if caching is enabled
text_hash = self._get_text_hash(text.strip())
if self.use_cache and text_hash in self.embedding_cache:
return self.embedding_cache[text_hash]
try:
response = openai.embeddings.create(
model="text-embedding-3-small",
input=text.strip()
)
embedding = np.array(response.data[0].embedding)
# Cache the embedding if caching is enabled
if self.use_cache:
self.embedding_cache[text_hash] = embedding
# Save cache periodically (every 10 new embeddings)
if len(self.embedding_cache) % 10 == 0:
self._save_embedding_cache()
return embedding
except Exception as e:
logging.error(f"Error getting embedding for '{text}': {e}")
# Return zero vector as fallback
return np.zeros(1536) # text-embedding-3-small dimension
def _get_embeddings_batch(self, texts: List[str], batch_name: str = "texts") -> List[np.ndarray]:
"""Get embeddings for multiple texts in batches with caching to improve performance"""
embeddings = []
texts_to_fetch = []
text_to_index = {}
start_time = time.time()
# Check cache for existing embeddings (if caching is enabled)
cache_hits = 0
for i, text in enumerate(texts):
if not text or not text.strip():
embeddings.append(None)
continue
text_hash = self._get_text_hash(text.strip())
if self.use_cache and text_hash in self.embedding_cache:
embeddings.append(self.embedding_cache[text_hash])
cache_hits += 1
else:
# Mark for fetching
embeddings.append(None) # Placeholder
texts_to_fetch.append(text.strip())
text_to_index[text.strip()] = i
cache_status = f"cache {'enabled' if self.use_cache else 'disabled'}"
logging.info(f"Computing embeddings for {len(texts)} {batch_name} ({cache_status}): {cache_hits} cache hits, {len(texts_to_fetch)} API calls needed")
if not texts_to_fetch:
logging.info(f"All embeddings found in cache!")
return embeddings
# Process remaining texts in batches
batch_size = 10
fetched_embeddings = {}
for i in range(0, len(texts_to_fetch), batch_size):
batch_start = time.time()
batch = texts_to_fetch[i:i+batch_size]
try:
batch_num = i//batch_size + 1
total_batches = (len(texts_to_fetch) + batch_size - 1)//batch_size
logging.info(f" Processing batch {batch_num}/{total_batches} ({len(batch)} texts)")
api_start = time.time()
response = openai.embeddings.create(
model="text-embedding-3-small",
input=batch
)
api_time = time.time() - api_start
for j, text in enumerate(batch):
embedding = np.array(response.data[j].embedding)
text_hash = self._get_text_hash(text)
# Cache the embedding if caching is enabled
if self.use_cache:
self.embedding_cache[text_hash] = embedding
fetched_embeddings[text] = embedding
batch_time = time.time() - batch_start
logging.info(f" Batch {batch_num} completed in {batch_time:.2f}s (API: {api_time:.2f}s)")
except Exception as e:
logging.error(f"Error getting embeddings for batch {batch_num}: {e}")
# Set None for failed texts
for text in batch:
fetched_embeddings[text] = None
# Fill in the fetched embeddings
for text, embedding in fetched_embeddings.items():
if text in text_to_index:
embeddings[text_to_index[text]] = embedding
# Save cache after batch processing (if caching is enabled)
if self.use_cache and len(texts_to_fetch) > 0:
self._save_embedding_cache()
total_time = time.time() - start_time
successful_count = len([e for e in embeddings if e is not None])
logging.info(f"Completed embedding computation for {batch_name}: {successful_count}/{len(embeddings)} successful in {total_time:.2f}s ({cache_hits} from cache)")
return embeddings
def _calculate_similarity_from_embeddings(self, emb1: np.ndarray, emb2: np.ndarray) -> float:
"""Calculate cosine similarity from precomputed embeddings"""
if emb1 is None or emb2 is None:
return 0.0
try:
# Reshape for sklearn
emb1 = emb1.reshape(1, -1)
emb2 = emb2.reshape(1, -1)
similarity = cosine_similarity(emb1, emb2)[0][0]
return float(similarity)
except Exception as e:
logging.error(f"Error calculating similarity from embeddings: {e}")
return 0.0
def calculate_similarity(self, text1: str, text2: str) -> float:
"""Calculate cosine similarity between two texts"""
emb1 = self.get_embedding(text1)
emb2 = self.get_embedding(text2)
# Reshape for sklearn
emb1 = emb1.reshape(1, -1)
emb2 = emb2.reshape(1, -1)
similarity = cosine_similarity(emb1, emb2)[0][0]
return float(similarity)
def compare_graphs(self, graph1_data: Dict[str, Any], graph2_data: Dict[str, Any]) -> GraphComparisonMetrics:
"""
Compare two knowledge graphs and generate comprehensive metrics.
Args:
graph1_data: First knowledge graph data
graph2_data: Second knowledge graph data
Returns:
Comprehensive comparison metrics
"""
start_time = time.time()
logging.info(f"Starting graph comparison at {time.strftime('%H:%M:%S')}")
# Extract entities and relations
entities1 = graph1_data.get('entities', [])
relations1 = graph1_data.get('relations', [])
entities2 = graph2_data.get('entities', [])
relations2 = graph2_data.get('relations', [])
# Check if graphs are from the same trace
trace_id1 = graph1_data.get('graph_info', {}).get('trace_id')
trace_id2 = graph2_data.get('graph_info', {}).get('trace_id')
kg_id1 = graph1_data.get('graph_info', {}).get('id')
kg_id2 = graph2_data.get('graph_info', {}).get('id')
is_same_trace = (trace_id1 and trace_id2 and trace_id1 == trace_id2 and kg_id1 != kg_id2)
logging.info(f"Graph comparison debug:")
logging.info(f" Graph 1 - trace_id: {trace_id1}, kg_id: {kg_id1}")
logging.info(f" Graph 2 - trace_id: {trace_id2}, kg_id: {kg_id2}")
logging.info(f" Same trace detected: {is_same_trace}")
logging.info(f" Will use {'SEMANTIC' if is_same_trace else 'EXACT'} comparison")
# Calculate entity metrics (use semantic comparison for same-trace graphs)
entity_start = time.time()
if is_same_trace:
logging.info("Using semantic entity comparison...")
entity_metrics = self._compare_entities_semantic(entities1, entities2)
else:
logging.info("Using exact entity comparison...")
entity_metrics = self._compare_entities(entities1, entities2)
entity_time = time.time() - entity_start
# Calculate relation metrics (use semantic comparison for same-trace graphs)
relation_start = time.time()
if is_same_trace:
logging.info("Using semantic relation comparison...")
relation_metrics = self._compare_relations_semantic(relations1, relations2)
else:
logging.info("Using exact relation comparison...")
relation_metrics = self._compare_relations(relations1, relations2)
relation_time = time.time() - relation_start
logging.info(f"Entity comparison results: overlap={entity_metrics['overlap_count']}, unique1={entity_metrics['unique_to_graph1']}, unique2={entity_metrics['unique_to_graph2']}")
logging.info(f"Relation comparison results: overlap={relation_metrics['overlap_count']}, unique1={relation_metrics['unique_to_graph1']}, unique2={relation_metrics['unique_to_graph2']}")
# Calculate structural metrics
structural_start = time.time()
structural_metrics = self._calculate_structural_metrics(entities1, relations1, entities2, relations2)
structural_time = time.time() - structural_start
# Calculate type distribution metrics
type_start = time.time()
type_metrics = self._calculate_type_distribution_metrics(entities1, relations1, entities2, relations2)
type_time = time.time() - type_start
# Calculate overall similarity scores
overall_start = time.time()
overall_metrics = self._calculate_overall_similarity(
entity_metrics, relation_metrics, structural_metrics, type_metrics
)
overall_time = time.time() - overall_start
# Generate graph statistics
stats_start = time.time()
graph1_stats = self._generate_graph_stats(entities1, relations1, "Graph 1")
graph2_stats = self._generate_graph_stats(entities2, relations2, "Graph 2")
stats_time = time.time() - stats_start
total_time = time.time() - start_time
logging.info(f"Graph comparison timing breakdown:")
logging.info(f" Entity comparison: {entity_time:.2f}s ({entity_time/total_time*100:.1f}%)")
logging.info(f" Relation comparison: {relation_time:.2f}s ({relation_time/total_time*100:.1f}%)")
logging.info(f" Structural metrics: {structural_time:.2f}s ({structural_time/total_time*100:.1f}%)")
logging.info(f" Type distribution: {type_time:.2f}s ({type_time/total_time*100:.1f}%)")
logging.info(f" Overall metrics: {overall_time:.2f}s ({overall_time/total_time*100:.1f}%)")
logging.info(f" Graph statistics: {stats_time:.2f}s ({stats_time/total_time*100:.1f}%)")
logging.info(f" TOTAL TIME: {total_time:.2f}s")
return GraphComparisonMetrics(
# Entity metrics
entity_overlap_count=entity_metrics['overlap_count'],
entity_unique_to_graph1=entity_metrics['unique_to_graph1'],
entity_unique_to_graph2=entity_metrics['unique_to_graph2'],
entity_overlap_ratio=entity_metrics['overlap_ratio'],
entity_semantic_similarity=entity_metrics['semantic_similarity'],
# Relation metrics
relation_overlap_count=relation_metrics['overlap_count'],
relation_unique_to_graph1=relation_metrics['unique_to_graph1'],
relation_unique_to_graph2=relation_metrics['unique_to_graph2'],
relation_overlap_ratio=relation_metrics['overlap_ratio'],
relation_semantic_similarity=relation_metrics['semantic_similarity'],
# Structural metrics
graph1_density=structural_metrics['graph1_density'],
graph2_density=structural_metrics['graph2_density'],
density_difference=structural_metrics['density_difference'],
common_patterns_count=structural_metrics['common_patterns_count'],
# Type distribution metrics
entity_type_similarity=type_metrics['entity_type_similarity'],
relation_type_similarity=type_metrics['relation_type_similarity'],
# Overall similarity scores
structural_similarity=overall_metrics['structural_similarity'],
content_similarity=overall_metrics['content_similarity'],
overall_similarity=overall_metrics['overall_similarity'],
# Additional statistics
graph1_stats=graph1_stats,
graph2_stats=graph2_stats
)
def _compare_entities(self, entities1: List[Dict], entities2: List[Dict]) -> Dict[str, Any]:
"""Compare entities between two graphs"""
# Create entity signatures for comparison
def create_entity_signature(entity):
return f"{entity.get('type', '')} {entity.get('name', '')}".strip().lower()
# Get entity sets
sig1_set = {create_entity_signature(e) for e in entities1}
sig2_set = {create_entity_signature(e) for e in entities2}
# Calculate overlap
overlap = sig1_set & sig2_set
unique_to_1 = sig1_set - sig2_set
unique_to_2 = sig2_set - sig1_set
# Calculate overlap ratio
total_unique = len(sig1_set | sig2_set)
overlap_ratio = len(overlap) / total_unique if total_unique > 0 else 0.0
# Calculate semantic similarity using embeddings
semantic_similarity = self._calculate_entity_semantic_similarity(entities1, entities2)
return {
'overlap_count': len(overlap),
'unique_to_graph1': len(unique_to_1),
'unique_to_graph2': len(unique_to_2),
'overlap_ratio': overlap_ratio,
'semantic_similarity': semantic_similarity
}
def _compare_relations(self, relations1: List[Dict], relations2: List[Dict]) -> Dict[str, Any]:
"""Compare relations between two graphs"""
# Create relation signatures for comparison
def create_relation_signature(relation):
return f"{relation.get('type', '')} {relation.get('description', '')}".strip().lower()
# Get relation sets
sig1_set = {create_relation_signature(r) for r in relations1}
sig2_set = {create_relation_signature(r) for r in relations2}
# Calculate overlap
overlap = sig1_set & sig2_set
unique_to_1 = sig1_set - sig2_set
unique_to_2 = sig2_set - sig1_set
# Calculate overlap ratio
total_unique = len(sig1_set | sig2_set)
overlap_ratio = len(overlap) / total_unique if total_unique > 0 else 0.0
# Calculate semantic similarity
semantic_similarity = self._calculate_relation_semantic_similarity(relations1, relations2)
return {
'overlap_count': len(overlap),
'unique_to_graph1': len(unique_to_1),
'unique_to_graph2': len(unique_to_2),
'overlap_ratio': overlap_ratio,
'semantic_similarity': semantic_similarity
}
def _calculate_entity_semantic_similarity(self, entities1: List[Dict], entities2: List[Dict]) -> float:
"""Calculate semantic similarity between entity sets using embeddings"""
if not entities1 or not entities2:
return 0.0
# Create text representations for entities
texts1 = [f"{e.get('type', '')} {e.get('name', '')} {e.get('description', '')}".strip() for e in entities1]
texts2 = [f"{e.get('type', '')} {e.get('name', '')} {e.get('description', '')}".strip() for e in entities2]
# Calculate similarity matrix
similarities = []
for text1 in texts1:
best_sim = 0.0
for text2 in texts2:
sim = self.calculate_similarity(text1, text2)
best_sim = max(best_sim, sim)
similarities.append(best_sim)
return np.mean(similarities) if similarities else 0.0
def _calculate_relation_semantic_similarity(self, relations1: List[Dict], relations2: List[Dict]) -> float:
"""Calculate semantic similarity between relation sets using embeddings"""
if not relations1 or not relations2:
return 0.0
# Create text representations for relations
texts1 = [f"{r.get('type', '')} {r.get('description', '')}".strip() for r in relations1]
texts2 = [f"{r.get('type', '')} {r.get('description', '')}".strip() for r in relations2]
# Calculate similarity matrix
similarities = []
for text1 in texts1:
best_sim = 0.0
for text2 in texts2:
sim = self.calculate_similarity(text1, text2)
best_sim = max(best_sim, sim)
similarities.append(best_sim)
return np.mean(similarities) if similarities else 0.0
def _calculate_structural_metrics(self, entities1: List[Dict], relations1: List[Dict],
entities2: List[Dict], relations2: List[Dict]) -> Dict[str, Any]:
"""Calculate structural similarity metrics"""
# Calculate graph densities
n1 = len(entities1)
e1 = len(relations1)
density1 = (2 * e1) / (n1 * (n1 - 1)) if n1 > 1 else 0.0
n2 = len(entities2)
e2 = len(relations2)
density2 = (2 * e2) / (n2 * (n2 - 1)) if n2 > 1 else 0.0
density_difference = abs(density1 - density2)
# Find common patterns (simple heuristic based on relation types)
pattern1 = self._extract_patterns(relations1)
pattern2 = self._extract_patterns(relations2)
common_patterns = len(set(pattern1) & set(pattern2))
return {
'graph1_density': density1,
'graph2_density': density2,
'density_difference': density_difference,
'common_patterns_count': common_patterns
}
def _extract_patterns(self, relations: List[Dict]) -> List[str]:
"""Extract structural patterns from relations"""
patterns = []
for relation in relations:
pattern = f"{relation.get('type', 'UNKNOWN')}"
patterns.append(pattern)
return patterns
def _calculate_type_distribution_metrics(self, entities1: List[Dict], relations1: List[Dict],
entities2: List[Dict], relations2: List[Dict]) -> Dict[str, Any]:
"""Calculate type distribution similarity metrics"""
# Entity type distributions
entity_types1 = {}
for entity in entities1:
etype = entity.get('type', 'Unknown')
entity_types1[etype] = entity_types1.get(etype, 0) + 1
entity_types2 = {}
for entity in entities2:
etype = entity.get('type', 'Unknown')
entity_types2[etype] = entity_types2.get(etype, 0) + 1
# Relation type distributions
relation_types1 = {}
for relation in relations1:
rtype = relation.get('type', 'Unknown')
relation_types1[rtype] = relation_types1.get(rtype, 0) + 1
relation_types2 = {}
for relation in relations2:
rtype = relation.get('type', 'Unknown')
relation_types2[rtype] = relation_types2.get(rtype, 0) + 1
# Calculate similarity using cosine similarity of type distributions
entity_type_similarity = self._calculate_distribution_similarity(entity_types1, entity_types2)
relation_type_similarity = self._calculate_distribution_similarity(relation_types1, relation_types2)
return {
'entity_type_similarity': entity_type_similarity,
'relation_type_similarity': relation_type_similarity
}
def _calculate_distribution_similarity(self, dist1: Dict[str, int], dist2: Dict[str, int]) -> float:
"""Calculate similarity between two distributions using cosine similarity"""
if not dist1 and not dist2:
return 1.0
if not dist1 or not dist2:
return 0.0
# Get all unique keys
all_keys = set(dist1.keys()) | set(dist2.keys())
# Create vectors
vec1 = np.array([dist1.get(key, 0) for key in all_keys])
vec2 = np.array([dist2.get(key, 0) for key in all_keys])
# Calculate cosine similarity
if np.sum(vec1) == 0 or np.sum(vec2) == 0:
return 0.0
vec1 = vec1.reshape(1, -1)
vec2 = vec2.reshape(1, -1)
similarity = cosine_similarity(vec1, vec2)[0][0]
return float(similarity)
def _calculate_overall_similarity(self, entity_metrics: Dict, relation_metrics: Dict,
structural_metrics: Dict, type_metrics: Dict) -> Dict[str, Any]:
"""Calculate overall similarity scores"""
# Structural similarity (combination of density and type distribution)
structural_sim = (
(1 - structural_metrics['density_difference']) * 0.3 +
type_metrics['entity_type_similarity'] * 0.35 +
type_metrics['relation_type_similarity'] * 0.35
)
# Content similarity (combination of entity and relation overlaps)
content_sim = (
entity_metrics['overlap_ratio'] * 0.4 +
relation_metrics['overlap_ratio'] * 0.3 +
entity_metrics['semantic_similarity'] * 0.15 +
relation_metrics['semantic_similarity'] * 0.15
)
# Overall similarity (weighted combination)
overall_sim = structural_sim * 0.4 + content_sim * 0.6
return {
'structural_similarity': max(0.0, min(1.0, structural_sim)),
'content_similarity': max(0.0, min(1.0, content_sim)),
'overall_similarity': max(0.0, min(1.0, overall_sim))
}
def _generate_graph_stats(self, entities: List[Dict], relations: List[Dict], graph_name: str) -> Dict[str, Any]:
"""Generate comprehensive statistics for a graph"""
# Entity type counts
entity_types = {}
for entity in entities:
etype = entity.get('type', 'Unknown')
entity_types[etype] = entity_types.get(etype, 0) + 1
# Relation type counts
relation_types = {}
for relation in relations:
rtype = relation.get('type', 'Unknown')
relation_types[rtype] = relation_types.get(rtype, 0) + 1
# Calculate basic metrics
n_entities = len(entities)
n_relations = len(relations)
density = (2 * n_relations) / (n_entities * (n_entities - 1)) if n_entities > 1 else 0.0
return {
'name': graph_name,
'entity_count': n_entities,
'relation_count': n_relations,
'density': density,
'entity_types': entity_types,
'relation_types': relation_types,
'avg_relations_per_entity': n_relations / n_entities if n_entities > 0 else 0.0
}
def _compare_entities_semantic(self, entities1: List[Dict], entities2: List[Dict]) -> Dict[str, Any]:
"""Compare entities using semantic similarity for overlap detection"""
if not entities1 or not entities2:
return {
'overlap_count': 0,
'unique_to_graph1': len(entities1),
'unique_to_graph2': len(entities2),
'overlap_ratio': 0.0,
'semantic_similarity': 0.0
}
logging.info(f"Starting semantic entity comparison: {len(entities1)} entities in graph1, {len(entities2)} entities in graph2")
logging.info(f"Total potential comparisons: {len(entities1)} x {len(entities2)} = {len(entities1) * len(entities2)}")
# Pre-compute all text representations
logging.info("Pre-computing text representations for entities...")
texts1 = []
texts2 = []
for i, entity1 in enumerate(entities1):
type1 = entity1.get('type', '').strip()
name1 = entity1.get('name', '').strip()
desc1 = entity1.get('description', '').strip()
text1_parts = [type1, name1]
if desc1:
text1_parts.append(desc1)
text1 = ' '.join(filter(None, text1_parts)).strip()
texts1.append(text1)
if i % 5 == 0 or i == len(entities1) - 1:
logging.info(f" Processed {i+1}/{len(entities1)} entities from graph1")
for i, entity2 in enumerate(entities2):
type2 = entity2.get('type', '').strip()
name2 = entity2.get('name', '').strip()
desc2 = entity2.get('description', '').strip()
text2_parts = [type2, name2]
if desc2:
text2_parts.append(desc2)
text2 = ' '.join(filter(None, text2_parts)).strip()
texts2.append(text2)
if i % 5 == 0 or i == len(entities2) - 1:
logging.info(f" Processed {i+1}/{len(entities2)} entities from graph2")
# Batch compute embeddings
logging.info("Computing embeddings in batches...")
embeddings1 = self._get_embeddings_batch(texts1, "graph1 entities")
embeddings2 = self._get_embeddings_batch(texts2, "graph2 entities")
# Find semantic matches using similarity threshold
logging.info(f"Finding semantic matches with threshold {self.semantic_threshold}...")
matched_entities1 = set()
matched_entities2 = set()
overlap_count = 0
total_comparisons = 0
for i, (entity1, text1, emb1) in enumerate(zip(entities1, texts1, embeddings1)):
if not text1 or emb1 is None: # Skip entities with no meaningful text or failed embeddings
continue
best_match_idx = None
best_similarity = 0.0
for j, (entity2, text2, emb2) in enumerate(zip(entities2, texts2, embeddings2)):
if j in matched_entities2 or not text2 or emb2 is None:
continue
# Calculate similarity using precomputed embeddings
similarity = self._calculate_similarity_from_embeddings(emb1, emb2)
total_comparisons += 1
if similarity >= self.semantic_threshold and similarity > best_similarity:
best_similarity = similarity
best_match_idx = j
if best_match_idx is not None:
matched_entities1.add(i)
matched_entities2.add(best_match_idx)
overlap_count += 1
logging.info(f" Match found: entity {i} ('{texts1[i][:50]}...') -> entity {best_match_idx} ('{texts2[best_match_idx][:50]}...'), similarity: {best_similarity:.3f}")
if i % 5 == 0 or i == len(entities1) - 1:
logging.info(f" Processed {i+1}/{len(entities1)} entities from graph1, found {overlap_count} matches so far")
unique_to_1 = len(entities1) - overlap_count
unique_to_2 = len(entities2) - overlap_count
# Calculate overlap ratio
total_unique = len(entities1) + len(entities2) - overlap_count
overlap_ratio = overlap_count / total_unique if total_unique > 0 else 0.0
# Calculate semantic similarity using existing method
semantic_similarity = self._calculate_entity_semantic_similarity(entities1, entities2)
logging.info(f"Entity semantic comparison completed:")
logging.info(f" Total comparisons made: {total_comparisons}")
logging.info(f" Overlaps found: {overlap_count}")
logging.info(f" Unique to graph1: {unique_to_1}")
logging.info(f" Unique to graph2: {unique_to_2}")
logging.info(f" Overlap ratio: {overlap_ratio:.3f}")
return {
'overlap_count': overlap_count,
'unique_to_graph1': unique_to_1,
'unique_to_graph2': unique_to_2,
'overlap_ratio': overlap_ratio,
'semantic_similarity': semantic_similarity
}
def _compare_relations_semantic(self, relations1: List[Dict], relations2: List[Dict]) -> Dict[str, Any]:
"""Compare relations using semantic similarity for overlap detection"""
if not relations1 or not relations2:
return {
'overlap_count': 0,
'unique_to_graph1': len(relations1),
'unique_to_graph2': len(relations2),
'overlap_ratio': 0.0,
'semantic_similarity': 0.0
}
logging.info(f"Starting semantic relation comparison: {len(relations1)} relations in graph1, {len(relations2)} relations in graph2")
logging.info(f"Total potential comparisons: {len(relations1)} x {len(relations2)} = {len(relations1) * len(relations2)}")
# Pre-compute all text representations
logging.info("Pre-computing text representations for relations...")
texts1 = []
texts2 = []
for i, relation1 in enumerate(relations1):
type1 = relation1.get('type', '').strip()
desc1 = relation1.get('description', '').strip()
source1 = relation1.get('source', '').strip()
target1 = relation1.get('target', '').strip()
text1_parts = [type1]
if desc1:
text1_parts.append(desc1)
if source1 and target1:
text1_parts.append(f"from {source1} to {target1}")
elif source1:
text1_parts.append(f"from {source1}")
elif target1:
text1_parts.append(f"to {target1}")
text1 = ' '.join(filter(None, text1_parts)).strip()
texts1.append(text1)
if i % 5 == 0 or i == len(relations1) - 1:
logging.info(f" Processed {i+1}/{len(relations1)} relations from graph1")
for i, relation2 in enumerate(relations2):
type2 = relation2.get('type', '').strip()
desc2 = relation2.get('description', '').strip()
source2 = relation2.get('source', '').strip()
target2 = relation2.get('target', '').strip()
text2_parts = [type2]
if desc2:
text2_parts.append(desc2)
if source2 and target2:
text2_parts.append(f"from {source2} to {target2}")
elif source2:
text2_parts.append(f"from {source2}")
elif target2:
text2_parts.append(f"to {target2}")
text2 = ' '.join(filter(None, text2_parts)).strip()
texts2.append(text2)
if i % 5 == 0 or i == len(relations2) - 1:
logging.info(f" Processed {i+1}/{len(relations2)} relations from graph2")
# Batch compute embeddings
logging.info("Computing embeddings in batches...")
embeddings1 = self._get_embeddings_batch(texts1, "graph1 relations")
embeddings2 = self._get_embeddings_batch(texts2, "graph2 relations")
# Find semantic matches using similarity threshold
logging.info(f"Finding semantic matches with threshold {self.semantic_threshold}...")
matched_relations1 = set()
matched_relations2 = set()
overlap_count = 0
total_comparisons = 0
for i, (relation1, text1, emb1) in enumerate(zip(relations1, texts1, embeddings1)):
if not text1 or emb1 is None: # Skip relations with no meaningful text or failed embeddings
continue
best_match_idx = None
best_similarity = 0.0
for j, (relation2, text2, emb2) in enumerate(zip(relations2, texts2, embeddings2)):
if j in matched_relations2 or not text2 or emb2 is None:
continue
# Calculate similarity using precomputed embeddings
similarity = self._calculate_similarity_from_embeddings(emb1, emb2)
total_comparisons += 1
if similarity >= self.semantic_threshold and similarity > best_similarity:
best_similarity = similarity
best_match_idx = j
if best_match_idx is not None:
matched_relations1.add(i)
matched_relations2.add(best_match_idx)
overlap_count += 1
logging.info(f" Match found: relation {i} ('{texts1[i][:50]}...') -> relation {best_match_idx} ('{texts2[best_match_idx][:50]}...'), similarity: {best_similarity:.3f}")
if i % 5 == 0 or i == len(relations1) - 1:
logging.info(f" Processed {i+1}/{len(relations1)} relations from graph1, found {overlap_count} matches so far")
unique_to_1 = len(relations1) - overlap_count
unique_to_2 = len(relations2) - overlap_count
# Calculate overlap ratio
total_unique = len(relations1) + len(relations2) - overlap_count
overlap_ratio = overlap_count / total_unique if total_unique > 0 else 0.0
# Calculate semantic similarity using existing method
semantic_similarity = self._calculate_relation_semantic_similarity(relations1, relations2)
logging.info(f"Relation semantic comparison completed:")
logging.info(f" Total comparisons made: {total_comparisons}")
logging.info(f" Overlaps found: {overlap_count}")
logging.info(f" Unique to graph1: {unique_to_1}")
logging.info(f" Unique to graph2: {unique_to_2}")
logging.info(f" Overlap ratio: {overlap_ratio:.3f}")
return {
'overlap_count': overlap_count,
'unique_to_graph1': unique_to_1,
'unique_to_graph2': unique_to_2,
'overlap_ratio': overlap_ratio,
'semantic_similarity': semantic_similarity
}
def compare_knowledge_graphs(graph1_data: Dict[str, Any], graph2_data: Dict[str, Any],
similarity_threshold: float = 0.7, semantic_threshold: float = 0.75,
use_cache: bool = True) -> GraphComparisonMetrics:
"""
Convenience function to compare two knowledge graphs.
Args:
graph1_data: First knowledge graph data
graph2_data: Second knowledge graph data
similarity_threshold: Threshold for semantic similarity matching (0.7 = 70%)
semantic_threshold: Threshold for semantic overlap detection (0.75 = 75%)
use_cache: Whether to use embedding cache (default: True)
Returns:
Comprehensive comparison metrics
"""
comparator = KnowledgeGraphComparator(
similarity_threshold=similarity_threshold,
semantic_threshold=semantic_threshold,
use_cache=use_cache
)
return comparator.compare_graphs(graph1_data, graph2_data) |