File size: 20,499 Bytes
f81cf0b |
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 |
"""
Hybrid search combining BM25 and vector similarity.
"""
from typing import List, Tuple, Optional, Dict, Any
import numpy as np
from django.db import connection
from django.db.models import QuerySet, F
from django.contrib.postgres.search import SearchQuery, SearchRank
from .embeddings import (
get_embedding_model,
generate_embedding,
cosine_similarity
)
from .embedding_utils import load_embedding
from .search_ml import expand_query_with_synonyms
# Default weights for hybrid search
DEFAULT_BM25_WEIGHT = 0.4
DEFAULT_VECTOR_WEIGHT = 0.6
# Minimum scores
DEFAULT_MIN_BM25_SCORE = 0.0
DEFAULT_MIN_VECTOR_SCORE = 0.1
def calculate_exact_match_boost(obj: Any, query: str, text_fields: List[str]) -> float:
"""
Calculate boost score for exact keyword matches in title/name fields.
Args:
obj: Django model instance.
query: Search query string.
text_fields: List of field names to check (first 2 are usually title/name).
Returns:
Boost score (0.0 to 1.0).
"""
if not query or not text_fields:
return 0.0
query_lower = query.lower().strip()
# Extract key phrases (2-3 word combinations) from query
query_words = query_lower.split()
key_phrases = []
for i in range(len(query_words) - 1):
phrase = " ".join(query_words[i:i+2])
if len(phrase) > 3:
key_phrases.append(phrase)
for i in range(len(query_words) - 2):
phrase = " ".join(query_words[i:i+3])
if len(phrase) > 5:
key_phrases.append(phrase)
# Also add individual words (longer than 2 chars)
query_words_set = set(word for word in query_words if len(word) > 2)
boost = 0.0
# Check primary fields (title, name) for exact matches
# First 2 fields are usually title/name
for field in text_fields[:2]:
if hasattr(obj, field):
field_value = str(getattr(obj, field, "")).lower()
if field_value:
# Check for key phrases first (highest priority)
for phrase in key_phrases:
if phrase in field_value:
# Major boost for phrase match
boost += 0.5
# Extra boost if it's the exact field value
if field_value.strip() == phrase.strip():
boost += 0.3
# Check for full query match
if query_lower in field_value:
boost += 0.4
# Count matched individual words
matched_words = sum(1 for word in query_words_set if word in field_value)
if matched_words > 0:
# Moderate boost for word matches
boost += 0.1 * min(matched_words, 3) # Cap at 3 words
return min(boost, 1.0) # Cap at 1.0 for very strong matches
def get_bm25_scores(
queryset: QuerySet,
query: str,
top_k: int = 20
) -> List[Tuple[Any, float]]:
"""
Get BM25 scores for queryset.
Args:
queryset: Django QuerySet to search.
query: Search query string.
top_k: Maximum number of results.
Returns:
List of (object, bm25_score) tuples.
"""
if not query or connection.vendor != "postgresql":
return []
if not hasattr(queryset.model, "tsv_body"):
return []
try:
expanded_queries = expand_query_with_synonyms(query)
combined_query = None
for q_variant in expanded_queries:
variant_query = SearchQuery(q_variant, config="simple")
combined_query = variant_query if combined_query is None else combined_query | variant_query
if combined_query is not None:
ranked_qs = (
queryset
.annotate(rank=SearchRank(F("tsv_body"), combined_query))
.filter(rank__gt=DEFAULT_MIN_BM25_SCORE)
.order_by("-rank")
)
results = list(ranked_qs[:top_k * 2]) # Get more for hybrid ranking
return [(obj, float(getattr(obj, "rank", 0.0))) for obj in results]
except Exception as e:
print(f"Error in BM25 search: {e}")
return []
def get_vector_scores(
queryset: QuerySet,
query: str,
top_k: int = 20
) -> List[Tuple[Any, float]]:
"""
Get vector similarity scores for queryset.
Args:
queryset: Django QuerySet to search.
query: Search query string.
top_k: Maximum number of results.
Returns:
List of (object, vector_score) tuples.
"""
if not query:
return []
# Generate query embedding
model = get_embedding_model()
if model is None:
return []
query_embedding = generate_embedding(query, model=model)
if query_embedding is None:
return []
# Get all objects with embeddings
all_objects = list(queryset)
if not all_objects:
return []
# Check dimension compatibility first
query_dim = len(query_embedding)
dimension_mismatch = False
# Calculate similarities
scores = []
for obj in all_objects:
obj_embedding = load_embedding(obj)
if obj_embedding is not None:
obj_dim = len(obj_embedding)
if obj_dim != query_dim:
# Dimension mismatch - skip vector search for this object
if not dimension_mismatch:
print(f"⚠️ Dimension mismatch: query={query_dim}, stored={obj_dim}. Skipping vector search.")
dimension_mismatch = True
continue
similarity = cosine_similarity(query_embedding, obj_embedding)
if similarity >= DEFAULT_MIN_VECTOR_SCORE:
scores.append((obj, similarity))
# If dimension mismatch detected, return empty to fall back to BM25 + exact match
if dimension_mismatch and not scores:
return []
# Sort by score descending
scores.sort(key=lambda x: x[1], reverse=True)
return scores[:top_k * 2] # Get more for hybrid ranking
def normalize_scores(scores: List[Tuple[Any, float]]) -> Dict[Any, float]:
"""
Normalize scores to 0-1 range.
Args:
scores: List of (object, score) tuples.
Returns:
Dictionary mapping object to normalized score.
"""
if not scores:
return {}
max_score = max(score for _, score in scores) if scores else 1.0
min_score = min(score for _, score in scores) if scores else 0.0
if max_score == min_score:
# All scores are the same, return uniform distribution
return {obj: 1.0 for obj, _ in scores}
# Normalize to 0-1
normalized = {}
for obj, score in scores:
normalized[obj] = (score - min_score) / (max_score - min_score)
return normalized
def hybrid_search(
queryset: QuerySet,
query: str,
top_k: int = 20,
bm25_weight: float = DEFAULT_BM25_WEIGHT,
vector_weight: float = DEFAULT_VECTOR_WEIGHT,
min_hybrid_score: float = 0.1,
text_fields: Optional[List[str]] = None
) -> List[Any]:
"""
Perform hybrid search combining BM25 and vector similarity.
Args:
queryset: Django QuerySet to search.
query: Search query string.
top_k: Maximum number of results.
bm25_weight: Weight for BM25 score (0-1).
vector_weight: Weight for vector score (0-1).
min_hybrid_score: Minimum combined score threshold.
text_fields: List of field names for exact match boost (optional).
Returns:
List of objects sorted by hybrid score.
"""
if not query:
return list(queryset[:top_k])
# Normalize weights
total_weight = bm25_weight + vector_weight
if total_weight > 0:
bm25_weight = bm25_weight / total_weight
vector_weight = vector_weight / total_weight
else:
bm25_weight = 0.5
vector_weight = 0.5
# Get BM25 scores
bm25_results = get_bm25_scores(queryset, query, top_k=top_k)
bm25_scores = normalize_scores(bm25_results)
# Get vector scores
vector_results = get_vector_scores(queryset, query, top_k=top_k)
vector_scores = normalize_scores(vector_results)
# Combine scores
combined_scores = {}
all_objects = set()
# Add BM25 objects
for obj, _ in bm25_results:
all_objects.add(obj)
combined_scores[obj] = bm25_scores.get(obj, 0.0) * bm25_weight
# Add vector objects
for obj, _ in vector_results:
all_objects.add(obj)
if obj in combined_scores:
combined_scores[obj] += vector_scores.get(obj, 0.0) * vector_weight
else:
combined_scores[obj] = vector_scores.get(obj, 0.0) * vector_weight
# CRITICAL: Find exact matches FIRST using icontains, then apply boost
# This ensures exact matches are always found and prioritized
if text_fields:
query_lower = query.lower()
# Extract key phrases (2-word and 3-word) from query
query_words = query_lower.split()
key_phrases = []
# 2-word phrases
for i in range(len(query_words) - 1):
phrase = " ".join(query_words[i:i+2])
if len(phrase) > 3:
key_phrases.append(phrase)
# 3-word phrases
for i in range(len(query_words) - 2):
phrase = " ".join(query_words[i:i+3])
if len(phrase) > 5:
key_phrases.append(phrase)
# Find potential exact matches using icontains on name/title field
# This ensures we don't miss exact matches even if BM25/vector don't find them
exact_match_candidates = set()
primary_field = text_fields[0] if text_fields else "name"
if hasattr(queryset.model, primary_field):
# Search for key phrases in the primary field
for phrase in key_phrases:
filter_kwargs = {f"{primary_field}__icontains": phrase}
candidates = queryset.filter(**filter_kwargs)[:top_k * 2]
exact_match_candidates.update(candidates)
# Apply exact match boost to all candidates
for obj in exact_match_candidates:
if obj not in all_objects:
all_objects.add(obj)
combined_scores[obj] = 0.0
# Apply exact match boost (this should dominate)
boost = calculate_exact_match_boost(obj, query, text_fields)
if boost > 0:
# Exact match boost should dominate - set it high
combined_scores[obj] = max(combined_scores.get(obj, 0.0), boost)
# Also check objects already in results for exact matches
for obj in list(all_objects):
boost = calculate_exact_match_boost(obj, query, text_fields)
if boost > 0:
# Boost existing scores
combined_scores[obj] = max(combined_scores.get(obj, 0.0), boost)
# Filter by minimum score and sort
filtered_scores = [
(obj, score) for obj, score in combined_scores.items()
if score >= min_hybrid_score
]
filtered_scores.sort(key=lambda x: x[1], reverse=True)
# Return top k
results = [obj for obj, _ in filtered_scores[:top_k]]
# Store hybrid score on objects for reference
for obj, score in filtered_scores[:top_k]:
obj._hybrid_score = score
obj._bm25_score = bm25_scores.get(obj, 0.0)
obj._vector_score = vector_scores.get(obj, 0.0)
# Store exact match boost if applied
if text_fields:
obj._exact_match_boost = calculate_exact_match_boost(obj, query, text_fields)
else:
obj._exact_match_boost = 0.0
return results
def semantic_query_expansion(query: str, top_n: int = 3) -> List[str]:
"""
Expand query with semantically similar terms using embeddings.
Args:
query: Original query string.
top_n: Number of similar terms to add.
Returns:
List of expanded query variations.
"""
try:
from hue_portal.chatbot.query_expansion import expand_query_semantically
return expand_query_semantically(query, context=None)
except Exception:
# Fallback to basic synonym expansion
return expand_query_with_synonyms(query)
def rerank_results(query: str, results: List[Any], text_fields: List[str], top_k: int = 5) -> List[Any]:
"""
Rerank results using cross-encoder approach (recalculate similarity with query).
Args:
query: Search query.
results: List of result objects.
text_fields: List of field names to use for reranking.
top_k: Number of top results to return.
Returns:
Reranked list of results.
"""
if not results or not query:
return results[:top_k]
try:
# Generate query embedding
model = get_embedding_model()
if model is None:
return results[:top_k]
query_embedding = generate_embedding(query, model=model)
if query_embedding is None:
return results[:top_k]
# Calculate similarity for each result
scored_results = []
for obj in results:
# Create text representation from text_fields
text_parts = []
for field in text_fields:
if hasattr(obj, field):
value = getattr(obj, field, "")
if value:
text_parts.append(str(value))
if not text_parts:
continue
obj_text = " ".join(text_parts)
obj_embedding = generate_embedding(obj_text, model=model)
if obj_embedding is not None:
similarity = cosine_similarity(query_embedding, obj_embedding)
scored_results.append((obj, similarity))
# Sort by similarity and return top_k
scored_results.sort(key=lambda x: x[1], reverse=True)
return [obj for obj, _ in scored_results[:top_k]]
except Exception as e:
print(f"Error in reranking: {e}")
return results[:top_k]
def diversify_results(results: List[Any], top_k: int = 5, similarity_threshold: float = 0.8) -> List[Any]:
"""
Ensure diversity in results by removing very similar items.
Args:
results: List of result objects.
top_k: Number of results to return.
similarity_threshold: Maximum similarity allowed between results.
Returns:
Diversified list of results.
"""
if len(results) <= top_k:
return results
try:
model = get_embedding_model()
if model is None:
return results[:top_k]
# Generate embeddings for all results
result_embeddings = []
valid_results = []
for obj in results:
# Try to get embedding from object
obj_embedding = load_embedding(obj)
if obj_embedding is not None:
result_embeddings.append(obj_embedding)
valid_results.append(obj)
if len(valid_results) <= top_k:
return valid_results
# Select diverse results using Maximal Marginal Relevance (MMR)
selected = [valid_results[0]] # Always include first (highest score)
selected_indices = {0}
selected_embeddings = [result_embeddings[0]]
for _ in range(min(top_k - 1, len(valid_results) - 1)):
best_score = -1
best_idx = -1
for i, (obj, emb) in enumerate(zip(valid_results, result_embeddings)):
if i in selected_indices:
continue
# Calculate max similarity to already selected results
max_sim = 0.0
for sel_emb in selected_embeddings:
sim = cosine_similarity(emb, sel_emb)
max_sim = max(max_sim, sim)
# Score: prefer results with lower similarity to selected ones
score = 1.0 - max_sim
if score > best_score:
best_score = score
best_idx = i
if best_idx >= 0:
selected.append(valid_results[best_idx])
selected_indices.add(best_idx)
selected_embeddings.append(result_embeddings[best_idx])
return selected
except Exception as e:
print(f"Error in diversifying results: {e}")
return results[:top_k]
def search_with_hybrid(
queryset: QuerySet,
query: str,
text_fields: List[str],
top_k: int = 20,
min_score: float = 0.1,
use_hybrid: bool = True,
bm25_weight: float = DEFAULT_BM25_WEIGHT,
vector_weight: float = DEFAULT_VECTOR_WEIGHT,
use_reranking: bool = False,
use_diversification: bool = False
) -> QuerySet:
"""
Search with hybrid BM25 + vector, with fallback to BM25-only or TF-IDF.
Args:
queryset: Django QuerySet to search.
query: Search query string.
text_fields: List of field names (for fallback).
top_k: Maximum number of results.
min_score: Minimum score threshold.
use_hybrid: Whether to use hybrid search.
bm25_weight: Weight for BM25 in hybrid search.
vector_weight: Weight for vector in hybrid search.
Returns:
Filtered and ranked QuerySet.
"""
if not query:
return queryset[:top_k]
# Try hybrid search if enabled
if use_hybrid:
try:
hybrid_results = hybrid_search(
queryset,
query,
top_k=top_k,
bm25_weight=bm25_weight,
vector_weight=vector_weight,
min_hybrid_score=min_score,
text_fields=text_fields
)
if hybrid_results:
# Apply reranking if enabled
if use_reranking and len(hybrid_results) > top_k:
hybrid_results = rerank_results(query, hybrid_results, text_fields, top_k=top_k * 2)
# Apply diversification if enabled
if use_diversification:
hybrid_results = diversify_results(hybrid_results, top_k=top_k)
# Convert to QuerySet with preserved order
result_ids = [obj.id for obj in hybrid_results[:top_k]]
if result_ids:
from django.db.models import Case, When, IntegerField
preserved = Case(
*[When(pk=pk, then=pos) for pos, pk in enumerate(result_ids)],
output_field=IntegerField()
)
return queryset.filter(id__in=result_ids).order_by(preserved)
except Exception as e:
print(f"Hybrid search failed, falling back: {e}")
# Fallback to BM25-only
if connection.vendor == "postgresql" and hasattr(queryset.model, "tsv_body"):
try:
expanded_queries = expand_query_with_synonyms(query)
combined_query = None
for q_variant in expanded_queries:
variant_query = SearchQuery(q_variant, config="simple")
combined_query = variant_query if combined_query is None else combined_query | variant_query
if combined_query is not None:
ranked_qs = (
queryset
.annotate(rank=SearchRank(F("tsv_body"), combined_query))
.filter(rank__gt=0)
.order_by("-rank")
)
results = list(ranked_qs[:top_k])
if results:
for obj in results:
obj._ml_score = getattr(obj, "rank", 0.0)
return results
except Exception:
pass
# Final fallback: import and use original search_with_ml
from .search_ml import search_with_ml
return search_with_ml(queryset, query, text_fields, top_k=top_k, min_score=min_score)
|