Spaces:
Running
Running
File size: 22,593 Bytes
0a4529c |
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 |
# DEPENDENCIES
import re
from typing import List
from typing import Optional
from config.models import DocumentChunk
from config.settings import get_settings
from config.models import DocumentMetadata
from config.models import ChunkingStrategy
from config.logging_config import get_logger
from chunking.base_chunker import BaseChunker
from chunking.base_chunker import ChunkerConfig
from chunking.token_counter import TokenCounter
from chunking.fixed_chunker import FixedChunker
from chunking.semantic_chunker import SemanticChunker
from chunking.llamaindex_chunker import LlamaIndexChunker
from chunking.hierarchical_chunker import HierarchicalChunker
# Setup Settings and Logging
logger = get_logger(__name__)
settings = get_settings()
class AdaptiveChunkingSelector:
"""
Intelligent chunking strategy selector with structure detection:
- Analyzes document characteristics (size, structure, content type)
- Detects structured documents (projects, sections, hierarchies)
- Automatically selects optimal chunking strategy
- Prioritizes section-aware chunking for structured content
Strategy Selection Logic (UPDATED):
- Small docs (< 1K tokens) β Fixed chunking
- Medium structured docs β Semantic (section-aware)
- Medium unstructured docs β LlamaIndex or basic semantic
- Large docs (>500K tokens) β Hierarchical chunking
"""
def __init__(self, prefer_llamaindex: bool = True):
"""
Initialize adaptive selector with all chunking strategies
Arguments:
----------
prefer_llamaindex { bool } : Prefer LlamaIndex over custom semantic chunking when available
"""
self.logger = logger
self.token_counter = TokenCounter()
self.prefer_llamaindex = prefer_llamaindex
# Initialize all chunking strategies
self.fixed_chunker = FixedChunker()
self.semantic_chunker = SemanticChunker(respect_section_boundaries = True)
self.hierarchical_chunker = HierarchicalChunker()
self.llamaindex_chunker = LlamaIndexChunker()
# Strategy thresholds (from settings)
self.small_doc_threshold = settings.SMALL_DOC_THRESHOLD
self.large_doc_threshold = settings.LARGE_DOC_THRESHOLD
# Check LlamaIndex availability
self.llamaindex_available = self.llamaindex_chunker._initialized
self.logger.info(f"Initialized AdaptiveChunkingSelector: LlamaIndex available={self.llamaindex_available}, prefer_llamaindex={self.prefer_llamaindex}, section_aware_semantic=True")
def select_chunking_strategy(self, text: str, metadata: Optional[DocumentMetadata] = None) -> tuple[ChunkingStrategy, dict]:
"""
Analyze document and select optimal chunking strategy: Detects structured documents and prioritizes section-aware chunking
Arguments:
----------
text { str } : Document text
metadata { DocumentMetadata } : Document metadata
Returns:
--------
{ tuple } : Tuple of (selected_strategy, analysis_results)
"""
analysis = self._analyze_document(text = text,
metadata = metadata,
)
# Check if document has clear structure (projects, sections)
has_structure = analysis.get("has_structure", False)
structure_score = analysis.get("structure_score", 0)
# Strategy selection logic
if (analysis["total_tokens"] <= self.small_doc_threshold):
strategy = ChunkingStrategy.FIXED
reason = f"Small document ({analysis['total_tokens']} tokens) - fixed chunking for simplicity"
elif (analysis["total_tokens"] <= self.large_doc_threshold):
# Medium documents: check for structure
if (has_structure and (structure_score > 0.3)):
# Structured document detected - use section-aware semantic chunking
strategy = ChunkingStrategy.SEMANTIC
reason = (f"Medium structured document ({analysis['total_tokens']} tokens, structure_score={structure_score:.2f}) - section-aware semantic chunking")
elif self.llamaindex_available and self.prefer_llamaindex:
strategy = ChunkingStrategy.SEMANTIC
reason = f"Medium document ({analysis['total_tokens']} tokens) - LlamaIndex semantic chunking"
else:
strategy = ChunkingStrategy.SEMANTIC
reason = f"Medium document ({analysis['total_tokens']} tokens) - semantic chunking"
else:
strategy = ChunkingStrategy.HIERARCHICAL
reason = f"Large document ({analysis['total_tokens']} tokens) - hierarchical chunking"
# Override based on document structure if available
if (metadata and self._has_clear_structure(metadata)):
if (strategy == ChunkingStrategy.FIXED):
# Upgrade to semantic for structured documents
strategy = ChunkingStrategy.SEMANTIC
reason = "Document has clear structure - section-aware semantic chunking preferred"
analysis["selected_strategy"] = strategy
analysis["selection_reason"] = reason
analysis["llamaindex_used"] = ((strategy == ChunkingStrategy.SEMANTIC) and self.llamaindex_available and self.prefer_llamaindex and not has_structure)
self.logger.info(f"Selected {strategy.value}: {reason}")
return strategy, analysis
def chunk_text(self, text: str, metadata: Optional[DocumentMetadata] = None, force_strategy: Optional[ChunkingStrategy] = None) -> List[DocumentChunk]:
"""
Automatically select strategy and chunk text
Arguments:
----------
text { str } : Document text
metadata { DocumentMetadata } : Document metadata
force_strategy { ChunkingStrategy } : Force specific strategy (optional)
Returns:
--------
{ list } : List of DocumentChunk objects
"""
if not text or not text.strip():
return []
# Select strategy (or use forced strategy)
if force_strategy:
strategy = force_strategy
analysis = self._analyze_document(text = text,
metadata = metadata,
)
reason = f"Forced strategy: {force_strategy.value}"
llamaindex_used = False
else:
strategy, analysis = self.select_chunking_strategy(text = text,
metadata = metadata,
)
reason = analysis["selection_reason"]
llamaindex_used = analysis["llamaindex_used"]
# Get appropriate chunker
if ((strategy == ChunkingStrategy.SEMANTIC) and llamaindex_used):
chunker = self.llamaindex_chunker
chunker_name = "LlamaIndex Semantic"
else:
chunker = self._get_chunker_for_strategy(strategy = strategy)
chunker_name = strategy.value
# Update metadata with strategy information
if metadata:
metadata.chunking_strategy = strategy
metadata.extra["chunking_analysis"] = {"strategy" : strategy.value,
"chunker_used" : chunker_name,
"reason" : reason,
"total_tokens" : analysis["total_tokens"],
"estimated_chunks" : analysis[f"estimated_{strategy.value.lower()}_chunks"],
"llamaindex_used" : llamaindex_used,
"has_structure" : analysis.get("has_structure", False),
"structure_score" : analysis.get("structure_score", 0),
}
self.logger.info(f"Using {chunker_name} chunker for document")
# Perform chunking
try:
chunks = chunker.chunk_text(text = text,
metadata = metadata,
)
# Add strategy metadata to chunks
for chunk in chunks:
chunk.metadata["chunking_strategy"] = strategy.value
chunk.metadata["chunker_used"] = chunker_name
if llamaindex_used:
chunk.metadata["llamaindex_splitter"] = self.llamaindex_chunker.splitter_type
self.logger.info(f"Successfully created {len(chunks)} chunks using {chunker_name}")
# Log section coverage statistics
chunks_with_sections = sum(1 for c in chunks if c.section_title)
if (chunks_with_sections > 0):
self.logger.info(f"Section coverage: {chunks_with_sections}/{len(chunks)} chunks ({chunks_with_sections/len(chunks)*100:.1f}%) have section titles")
return chunks
except Exception as e:
self.logger.error(f"{chunker_name} chunking failed: {repr(e)}, falling back to fixed chunking")
# Fallback to fixed chunking
return self.fixed_chunker.chunk_text(text = text,
metadata = metadata,
)
def _analyze_document(self, text: str, metadata: Optional[DocumentMetadata] = None) -> dict:
"""
Analyze document characteristics for strategy selection: Includes structure detection
Arguments:
----------
text { str } : Document text
metadata { DocumentMetadata } : Document metadata
Returns:
--------
{ dict } : Analysis results
"""
# Basic token analysis
total_tokens = self.token_counter.count_tokens(text = text)
total_chars = len(text)
total_words = len(text.split())
# Estimate chunks for each strategy
estimated_fixed_chunks = max(1, total_tokens // settings.FIXED_CHUNK_SIZE)
estimated_semantic_chunks = max(1, total_tokens // (settings.FIXED_CHUNK_SIZE * 2))
estimated_hierarchical_chunks = max(1, total_tokens // settings.CHILD_CHUNK_SIZE)
estimated_llamaindex_chunks = max(1, total_tokens // (settings.FIXED_CHUNK_SIZE * 1.5))
# Structure analysis (simple heuristics)
sentence_count = len(self.token_counter._split_into_sentences(text = text))
avg_sentence_length = total_words / sentence_count if (sentence_count > 0) else 0
# Paragraph detection (rough)
paragraphs = [p for p in text.split('\n\n') if p.strip()]
paragraph_count = len(paragraphs)
# NEW: Detect document structure
has_structure, structure_score = self._detect_document_structure(text)
analysis = {"total_tokens" : total_tokens,
"total_chars" : total_chars,
"total_words" : total_words,
"sentence_count" : sentence_count,
"paragraph_count" : paragraph_count,
"avg_sentence_length" : avg_sentence_length,
"estimated_fixed_chunks" : estimated_fixed_chunks,
"estimated_semantic_chunks" : estimated_semantic_chunks,
"estimated_llamaindex_chunks" : estimated_llamaindex_chunks,
"estimated_hierarchical_chunks" : estimated_hierarchical_chunks,
"document_size_category" : self._get_size_category(total_tokens),
"llamaindex_available" : self.llamaindex_available,
"has_structure" : has_structure,
"structure_score" : structure_score,
}
# Add metadata-based insights if available
if metadata:
analysis.update({"document_type" : metadata.document_type.value,
"file_size_mb" : metadata.file_size_mb,
"num_pages" : metadata.num_pages,
"has_clear_structure" : self._has_clear_structure(metadata),
})
return analysis
def _detect_document_structure(self, text: str) -> tuple[bool, float]:
"""
Analyzes text for structural patterns and detect if document has clear structural elements (projects, sections, etc.)
& returns: (has_structure, structure_score)
"""
structure_indicators = 0
max_indicators = 5
# Check for project-style headers: "a) Project Name", "b) Project Name"
project_headers = len(re.findall(r'^[a-z]\)\s+[A-Z]', text, re.MULTILINE))
if (project_headers > 2):
structure_indicators += 1
# Check for bullet point lists: "β" or "β"
bullet_points = text.count('β') + text.count('β')
if (bullet_points > 5):
structure_indicators += 1
# Check for numbered sections: "1.", "2.", etc.
numbered_sections = len(re.findall(r'^\d+\.\s+[A-Z]', text, re.MULTILINE))
if (numbered_sections > 2):
structure_indicators += 1
# Check for subsection markers ending with ":"
subsection_markers = len(re.findall(r'^β\s+\w+.*:', text, re.MULTILINE))
if (subsection_markers > 3):
structure_indicators += 1
# Check for consistent indentation patterns
lines = text.split('\n')
indented_lines = sum(1 for line in lines if line.startswith(' ') or line.startswith('\t'))
# >20% indented
if (indented_lines > len(lines) * 0.2):
structure_indicators += 1
has_structure = (structure_indicators >= 2)
structure_score = structure_indicators / max_indicators
if has_structure:
self.logger.info(f"Document structure detected: score={structure_score:.2f} (project_headers={project_headers}, bullets={bullet_points}, "
f"numbered_sections={numbered_sections}, subsections={subsection_markers})")
return has_structure, structure_score
def _get_chunker_for_strategy(self, strategy: ChunkingStrategy) -> BaseChunker:
"""
Get chunker instance for specified strategy
Arguments:
----------
strategy { ChunkingStrategy } : Chunking strategy
Returns:
--------
{ BaseChunker } : Chunker instance
"""
chunkers = {ChunkingStrategy.FIXED : self.fixed_chunker,
ChunkingStrategy.SEMANTIC : self.semantic_chunker,
ChunkingStrategy.HIERARCHICAL : self.hierarchical_chunker,
}
return chunkers.get(strategy, self.fixed_chunker)
def _get_size_category(self, total_tokens: int) -> str:
"""
Categorize document by size
"""
if (total_tokens <= self.small_doc_threshold):
return "small"
elif (total_tokens <= self.large_doc_threshold):
return "medium"
else:
return "large"
def _has_clear_structure(self, metadata: DocumentMetadata) -> bool:
"""
Check if document has clear structural elements
"""
if metadata.extra:
# DOCX with multiple sections/headings
if (metadata.document_type.value == "docx"):
if (metadata.extra.get("num_sections", 0) > 1):
return True
if (metadata.extra.get("num_paragraphs", 0) > 50):
return True
# PDF with multiple pages and likely structure
if (metadata.document_type.value == "pdf"):
if metadata.num_pages and metadata.num_pages > 10:
return True
return False
def get_strategy_recommendations(self, text: str, metadata: Optional[DocumentMetadata] = None) -> dict:
"""
Get detailed strategy recommendations with pros/cons
"""
analysis = self._analyze_document(text, metadata)
# LlamaIndex recommendation
llamaindex_recommendation = {"recommended_for" : ["Medium documents", "Structured content", "Superior semantic analysis"],
"pros" : ["Best semantic boundary detection", "LlamaIndex ecosystem integration", "Advanced embedding-based splitting"],
"cons" : ["Additional dependency", "Slower initialization", "More complex setup"],
"estimated_chunks" : analysis["estimated_llamaindex_chunks"],
"available" : self.llamaindex_available,
}
recommendations = {"fixed" : {"recommended_for" : ["Small documents", "Homogeneous content", "Simple processing"],
"pros" : ["Fast", "Reliable", "Predictable chunk sizes"],
"cons" : ["May break semantic boundaries", "Ignores document structure"],
"estimated_chunks" : analysis["estimated_fixed_chunks"],
},
"semantic" : {"recommended_for" : ["Medium documents", "Structured content", "When coherence matters"],
"pros" : ["Preserves topic boundaries", "Respects section structure", "Better context coherence"],
"cons" : ["Slower (requires embeddings)", "Less predictable chunk sizes"],
"estimated_chunks" : analysis["estimated_semantic_chunks"],
"section_aware" : True,
},
"llamaindex" : llamaindex_recommendation,
"hierarchical" : {"recommended_for" : ["Large documents", "Complex structure", "Granular search needs"],
"pros" : ["Best for large docs", "Granular + context search", "Scalable"],
"cons" : ["Complex implementation", "More chunks to manage", "Higher storage"],
"estimated_chunks" : analysis["estimated_hierarchical_chunks"],
}
}
# Add selected strategy
selected_strategy, analysis_result = self.select_chunking_strategy(text = text,
metadata = metadata,
)
recommendations["selected_strategy"] = selected_strategy.value
recommendations["selection_reason"] = analysis_result["selection_reason"]
recommendations["llamaindex_used"] = analysis_result["llamaindex_used"]
recommendations["structure_detected"] = analysis_result.get("has_structure", False)
return recommendations
# Global adaptive selector instance
_adaptive_selector = None
def get_adaptive_selector() -> AdaptiveChunkingSelector:
"""
Get global adaptive selector instance (singleton)
"""
global _adaptive_selector
if _adaptive_selector is None:
_adaptive_selector = AdaptiveChunkingSelector()
return _adaptive_selector
def adaptive_chunk_text(text: str, metadata: Optional[DocumentMetadata] = None, force_strategy: Optional[ChunkingStrategy] = None) -> List[DocumentChunk]:
"""
Convenience function for adaptive chunking
"""
selector = get_adaptive_selector()
return selector.chunk_text(text, metadata, force_strategy)
def analyze_document(text: str, metadata: Optional[DocumentMetadata] = None) -> dict:
"""
Analyze document without chunking
"""
selector = get_adaptive_selector()
return selector._analyze_document(text, metadata) |