Spaces:
Running
on
Zero
Running
on
Zero
File size: 25,515 Bytes
575f1c7 63279a9 575f1c7 63279a9 |
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 |
"""Text chunking strategies for RAG document processing."""
import re
from typing import List, Dict, Any, Tuple
from langchain_text_splitters import RecursiveCharacterTextSplitter
from langchain_core.documents import Document
from src.core.logging_config import get_logger
logger = get_logger(__name__)
class LaTeXAwareChunker:
"""Handles LaTeX-aware document chunking that preserves LaTeX structures."""
def __init__(self, chunk_size: int = 1200, chunk_overlap: int = 150):
"""
Initialize the LaTeX-aware document chunker.
Args:
chunk_size: Maximum size of each chunk in characters
chunk_overlap: Number of characters to overlap between chunks
"""
self.chunk_size = chunk_size
self.chunk_overlap = chunk_overlap
# Initialize the text splitter with LaTeX-aware settings
self.text_splitter = RecursiveCharacterTextSplitter(
chunk_size=chunk_size,
chunk_overlap=chunk_overlap,
length_function=len,
separators=[
"\n\\section{", # Section headers
"\n\\subsection{", # Subsection headers
"\n\\subsubsection{", # Subsubsection headers
"\n\\title{", # Title commands
"\n\\begin{", # Begin environments
"\n\\end{", # End environments
"\n\n", # Paragraph breaks
"\n", # Line breaks
". ", # Sentence breaks
" ", # Word breaks
"" # Character breaks
],
keep_separator=True,
add_start_index=True
)
# Regex patterns for LaTeX structures
self.latex_table_pattern = re.compile(
r'\\begin\{tabular\}.*?\\end\{tabular\}',
re.DOTALL | re.MULTILINE
)
self.latex_title_pattern = re.compile(
r'\\title\{[^}]*\}',
re.MULTILINE
)
self.latex_section_pattern = re.compile(
r'\\(?:sub)*section\*?\{[^}]*\}',
re.MULTILINE
)
self.latex_environment_pattern = re.compile(
r'\\begin\{[^}]+\}.*?\\end\{[^}]+\}',
re.DOTALL | re.MULTILINE
)
logger.info(f"LaTeX-aware chunker initialized with chunk_size={chunk_size}, overlap={chunk_overlap}")
def extract_latex_structures(self, content: str) -> Tuple[List[Tuple[int, int, str]], str]:
"""
Extract LaTeX tables and environments, replacing them with placeholders.
Args:
content: Original LaTeX content
Returns:
Tuple of (structures_list, content_with_placeholders)
"""
structures = []
# Find all tabular environments (highest priority)
for match in self.latex_table_pattern.finditer(content):
structures.append((
match.start(),
match.end(),
"latex_table",
match.group()
))
# Find other LaTeX environments (avoid overlapping with tables)
for match in self.latex_environment_pattern.finditer(content):
# Check if this environment overlaps with any table
overlaps_with_table = any(
table_start <= match.start() < table_end or
table_start < match.end() <= table_end
for table_start, table_end, struct_type, _ in structures
if struct_type == "latex_table"
)
if not overlaps_with_table and "tabular" not in match.group():
structures.append((
match.start(),
match.end(),
"latex_environment",
match.group()
))
# Find titles and sections
for match in self.latex_title_pattern.finditer(content):
structures.append((
match.start(),
match.end(),
"latex_title",
match.group()
))
for match in self.latex_section_pattern.finditer(content):
structures.append((
match.start(),
match.end(),
"latex_section",
match.group()
))
# Sort by start position
structures.sort(key=lambda x: x[0])
# Replace structures with placeholders
content_with_placeholders = content
offset = 0
for i, (start, end, struct_type, struct_content) in enumerate(structures):
placeholder = f"\n\n__LATEX_STRUCTURE_{i}_{struct_type.upper()}__\n\n"
# Adjust positions based on previous replacements
adjusted_start = start - offset
adjusted_end = end - offset
content_with_placeholders = (
content_with_placeholders[:adjusted_start] +
placeholder +
content_with_placeholders[adjusted_end:]
)
# Update offset for next replacement
offset += (end - start) - len(placeholder)
return structures, content_with_placeholders
def restore_latex_structures(self, chunks: List[str], structures: List[Tuple[int, int, str, str]]) -> List[str]:
"""
Restore LaTeX structures in chunks, keeping tables and environments intact.
Args:
chunks: List of text chunks with placeholders
structures: List of original structures
Returns:
List of chunks with restored structures
"""
restored_chunks = []
for chunk in chunks:
restored_chunk = chunk
# Find placeholders in this chunk
placeholder_pattern = re.compile(r'__LATEX_STRUCTURE_(\d+)_(\w+)__')
for match in placeholder_pattern.finditer(chunk):
structure_index = int(match.group(1))
if structure_index < len(structures):
original_structure = structures[structure_index][3]
restored_chunk = restored_chunk.replace(match.group(), original_structure)
restored_chunks.append(restored_chunk)
return restored_chunks
def chunk_document(self, content: str, source_metadata: Dict[str, Any]) -> List[Document]:
"""
Chunk a LaTeX document while preserving LaTeX structures.
Args:
content: The LaTeX content to chunk
source_metadata: Metadata about the source document
Returns:
List of Document objects with chunked content and enhanced metadata
"""
try:
# Extract LaTeX structures and replace with placeholders
structures, content_with_placeholders = self.extract_latex_structures(content)
# Create a document object with placeholders
doc = Document(
page_content=content_with_placeholders,
metadata=source_metadata
)
# Split the document into chunks
chunks = self.text_splitter.split_documents([doc])
# Restore LaTeX structures in chunks
chunk_contents = [chunk.page_content for chunk in chunks]
restored_contents = self.restore_latex_structures(chunk_contents, structures)
# Create enhanced chunks with restored content
enhanced_chunks = []
for i, (chunk, restored_content) in enumerate(zip(chunks, restored_contents)):
# Add chunk-specific metadata
chunk.metadata.update({
"chunk_index": i,
"total_chunks": len(chunks),
"chunk_size": len(restored_content),
"chunk_id": f"{source_metadata.get('source_id', 'unknown')}_{i}",
"has_latex_table": "\\begin{tabular}" in restored_content,
"has_latex_environment": "\\begin{" in restored_content and "\\end{" in restored_content,
"has_latex_math": "\\(" in restored_content or "$" in restored_content,
"content_type": "latex"
})
# Update the chunk content with restored structures
chunk.page_content = restored_content
enhanced_chunks.append(chunk)
logger.info(f"LaTeX document chunked into {len(enhanced_chunks)} structure-aware pieces")
return enhanced_chunks
except Exception as e:
logger.error(f"Error chunking LaTeX document: {e}")
# Fallback to regular chunking if LaTeX processing fails
return self._fallback_chunk(content, source_metadata)
def _fallback_chunk(self, content: str, source_metadata: Dict[str, Any]) -> List[Document]:
"""Fallback chunking method if LaTeX-aware chunking fails."""
try:
doc = Document(page_content=content, metadata=source_metadata)
chunks = self.text_splitter.split_documents([doc])
for i, chunk in enumerate(chunks):
chunk.metadata.update({
"chunk_index": i,
"total_chunks": len(chunks),
"chunk_size": len(chunk.page_content),
"chunk_id": f"{source_metadata.get('source_id', 'unknown')}_{i}",
"content_type": "latex"
})
logger.warning(f"Used fallback chunking for LaTeX content: {len(chunks)} pieces")
return chunks
except Exception as e:
logger.error(f"Error in LaTeX fallback chunking: {e}")
raise
class MarkdownAwareChunker:
"""Handles markdown-aware document chunking that preserves tables and structures."""
def __init__(self, chunk_size: int = 1000, chunk_overlap: int = 200):
"""
Initialize the markdown-aware document chunker.
Args:
chunk_size: Maximum size of each chunk in characters
chunk_overlap: Number of characters to overlap between chunks
"""
self.chunk_size = chunk_size
self.chunk_overlap = chunk_overlap
# Initialize the text splitter with markdown-aware settings
self.text_splitter = RecursiveCharacterTextSplitter(
chunk_size=chunk_size,
chunk_overlap=chunk_overlap,
length_function=len,
separators=[
"\n\n", # Paragraphs and sections
"\n# ", # H1 headers
"\n## ", # H2 headers
"\n### ", # H3 headers
"\n\n---\n\n", # Horizontal rules
"\n", # Lines
" ", # Words
".", # Sentences
",", # Clauses
"" # Characters
],
keep_separator=True,
add_start_index=True
)
# Regex patterns for markdown structures
self.table_pattern = re.compile(
r'(\|.*\|.*\n)+(\|[-\s|:]+\|.*\n)(\|.*\|.*\n)*',
re.MULTILINE
)
self.code_block_pattern = re.compile(
r'```[\s\S]*?```',
re.MULTILINE
)
logger.info(f"Markdown-aware chunker initialized with chunk_size={chunk_size}, overlap={chunk_overlap}")
def extract_markdown_structures(self, content: str) -> Tuple[List[Tuple[int, int, str]], str]:
"""
Extract markdown tables and code blocks, replacing them with placeholders.
Args:
content: Original markdown content
Returns:
Tuple of (structures_list, content_with_placeholders)
where structures_list contains (start, end, type, content) tuples
"""
structures = []
# Find all tables
for match in self.table_pattern.finditer(content):
structures.append((
match.start(),
match.end(),
"table",
match.group()
))
# Find all code blocks
for match in self.code_block_pattern.finditer(content):
structures.append((
match.start(),
match.end(),
"code_block",
match.group()
))
# Sort by start position
structures.sort(key=lambda x: x[0])
# Replace structures with placeholders
content_with_placeholders = content
offset = 0
for i, (start, end, struct_type, struct_content) in enumerate(structures):
placeholder = f"\n\n__STRUCTURE_{i}_{struct_type.upper()}__\n\n"
# Adjust positions based on previous replacements
adjusted_start = start - offset
adjusted_end = end - offset
content_with_placeholders = (
content_with_placeholders[:adjusted_start] +
placeholder +
content_with_placeholders[adjusted_end:]
)
# Update offset for next replacement
offset += (end - start) - len(placeholder)
return structures, content_with_placeholders
def restore_structures(self, chunks: List[str], structures: List[Tuple[int, int, str, str]]) -> List[str]:
"""
Restore markdown structures in chunks, keeping tables and code blocks intact.
Args:
chunks: List of text chunks with placeholders
structures: List of original structures
Returns:
List of chunks with restored structures
"""
restored_chunks = []
for chunk in chunks:
restored_chunk = chunk
# Find placeholders in this chunk
placeholder_pattern = re.compile(r'__STRUCTURE_(\d+)_(\w+)__')
for match in placeholder_pattern.finditer(chunk):
structure_index = int(match.group(1))
if structure_index < len(structures):
original_structure = structures[structure_index][3]
restored_chunk = restored_chunk.replace(match.group(), original_structure)
restored_chunks.append(restored_chunk)
return restored_chunks
def chunk_document(self, content: str, source_metadata: Dict[str, Any]) -> List[Document]:
"""
Chunk a markdown document while preserving tables and code blocks.
Args:
content: The markdown content to chunk
source_metadata: Metadata about the source document
Returns:
List of Document objects with chunked content and enhanced metadata
"""
try:
# Extract markdown structures (tables, code blocks) and replace with placeholders
structures, content_with_placeholders = self.extract_markdown_structures(content)
# Create a document object with placeholders
doc = Document(
page_content=content_with_placeholders,
metadata=source_metadata
)
# Split the document into chunks
chunks = self.text_splitter.split_documents([doc])
# Restore markdown structures in chunks
chunk_contents = [chunk.page_content for chunk in chunks]
restored_contents = self.restore_structures(chunk_contents, structures)
# Create enhanced chunks with restored content
enhanced_chunks = []
for i, (chunk, restored_content) in enumerate(zip(chunks, restored_contents)):
# Add chunk-specific metadata
chunk.metadata.update({
"chunk_index": i,
"total_chunks": len(chunks),
"chunk_size": len(restored_content),
"chunk_id": f"{source_metadata.get('source_id', 'unknown')}_{i}",
"has_table": "table" in restored_content.lower() and "|" in restored_content,
"has_code": "```" in restored_content
})
# Update the chunk content with restored structures
chunk.page_content = restored_content
enhanced_chunks.append(chunk)
logger.info(f"Document chunked into {len(enhanced_chunks)} markdown-aware pieces")
return enhanced_chunks
except Exception as e:
logger.error(f"Error chunking markdown document: {e}")
# Fallback to regular chunking if markdown processing fails
return self._fallback_chunk(content, source_metadata)
def _fallback_chunk(self, content: str, source_metadata: Dict[str, Any]) -> List[Document]:
"""Fallback chunking method if markdown-aware chunking fails."""
try:
doc = Document(page_content=content, metadata=source_metadata)
chunks = self.text_splitter.split_documents([doc])
for i, chunk in enumerate(chunks):
chunk.metadata.update({
"chunk_index": i,
"total_chunks": len(chunks),
"chunk_size": len(chunk.page_content),
"chunk_id": f"{source_metadata.get('source_id', 'unknown')}_{i}"
})
logger.warning(f"Used fallback chunking for {len(chunks)} pieces")
return chunks
except Exception as e:
logger.error(f"Error in fallback chunking: {e}")
raise
def chunk_multiple_documents(self, documents: List[Dict[str, Any]]) -> List[Document]:
"""
Chunk multiple documents for batch processing.
Args:
documents: List of dictionaries with 'content' and 'metadata' keys
Returns:
List of chunked Document objects
"""
all_chunks = []
for doc_data in documents:
content = doc_data.get('content', '')
metadata = doc_data.get('metadata', {})
if content.strip(): # Only process non-empty content
chunks = self.chunk_document(content, metadata)
all_chunks.extend(chunks)
logger.info(f"Chunked {len(documents)} documents into {len(all_chunks)} total chunks")
return all_chunks
def get_chunk_preview(self, chunks: List[Document], max_chunks: int = 5) -> str:
"""
Generate a preview of chunks for debugging/logging.
Args:
chunks: List of Document chunks
max_chunks: Maximum number of chunks to include in preview
Returns:
String preview of chunks
"""
preview = f"Document Chunks Preview ({len(chunks)} total chunks):\n"
preview += "=" * 50 + "\n"
for i, chunk in enumerate(chunks[:max_chunks]):
has_table = chunk.metadata.get('has_table', False)
has_code = chunk.metadata.get('has_code', False)
preview += f"Chunk {i + 1}:\n"
preview += f" Length: {len(chunk.page_content)} characters\n"
preview += f" Has Table: {has_table}, Has Code: {has_code}\n"
preview += f" Metadata: {chunk.metadata}\n"
preview += f" Content preview: {chunk.page_content[:100]}...\n"
preview += "-" * 30 + "\n"
if len(chunks) > max_chunks:
preview += f"... and {len(chunks) - max_chunks} more chunks\n"
return preview
class UnifiedDocumentChunker:
"""Unified chunker that handles both Markdown and LaTeX content types."""
def __init__(self):
"""Initialize the unified chunker with both markdown and LaTeX chunkers."""
self.markdown_chunker = MarkdownAwareChunker(chunk_size=1000, chunk_overlap=200)
self.latex_chunker = LaTeXAwareChunker(chunk_size=1200, chunk_overlap=150)
logger.info("Unified document chunker initialized with both Markdown and LaTeX support")
def chunk_document(self, content: str, source_metadata: Dict[str, Any]) -> List[Document]:
"""
Chunk a document using the appropriate chunker based on content type.
Args:
content: The document content to chunk
source_metadata: Metadata about the source document
Returns:
List of Document objects with chunked content and enhanced metadata
"""
# Determine content type from metadata or content analysis
content_type = source_metadata.get('doc_type', 'markdown').lower()
# Override content type detection for GOT-OCR results
if source_metadata.get('conversion_method', '').startswith('GOT-OCR'):
content_type = 'latex'
# Auto-detect content type if not specified
if content_type not in ['markdown', 'latex']:
if self._is_latex_content(content):
content_type = 'latex'
else:
content_type = 'markdown'
# Use appropriate chunker
if content_type == 'latex':
logger.info("Using LaTeX-aware chunker for document")
return self.latex_chunker.chunk_document(content, source_metadata)
else:
logger.info("Using Markdown-aware chunker for document")
return self.markdown_chunker.chunk_document(content, source_metadata)
def _is_latex_content(self, content: str) -> bool:
"""
Auto-detect if content is LaTeX based on common LaTeX commands.
Args:
content: Content to analyze
Returns:
True if content appears to be LaTeX, False otherwise
"""
latex_indicators = [
r'\\begin\{',
r'\\end\{',
r'\\title\{',
r'\\section',
r'\\subsection',
r'\\hline',
r'\\multirow',
r'\\multicolumn'
]
# Count LaTeX indicators
latex_count = sum(1 for indicator in latex_indicators if re.search(indicator, content))
# If we find multiple LaTeX indicators, treat as LaTeX
return latex_count >= 2
def chunk_multiple_documents(self, documents: List[Dict[str, Any]]) -> List[Document]:
"""
Chunk multiple documents using appropriate chunkers.
Args:
documents: List of dictionaries with 'content' and 'metadata' keys
Returns:
List of chunked Document objects
"""
all_chunks = []
for doc_data in documents:
content = doc_data.get('content', '')
metadata = doc_data.get('metadata', {})
if content.strip(): # Only process non-empty content
chunks = self.chunk_document(content, metadata)
all_chunks.extend(chunks)
logger.info(f"Chunked {len(documents)} documents into {len(all_chunks)} total chunks")
return all_chunks
def get_chunk_preview(self, chunks: List[Document], max_chunks: int = 5) -> str:
"""
Generate a preview of chunks for debugging/logging.
Args:
chunks: List of Document chunks
max_chunks: Maximum number of chunks to include in preview
Returns:
String preview of chunks
"""
preview = f"Document Chunks Preview ({len(chunks)} total chunks):\n"
preview += "=" * 50 + "\n"
for i, chunk in enumerate(chunks[:max_chunks]):
content_type = chunk.metadata.get('content_type', 'unknown')
has_table = chunk.metadata.get('has_table', False) or chunk.metadata.get('has_latex_table', False)
has_code = chunk.metadata.get('has_code', False) or chunk.metadata.get('has_latex_environment', False)
preview += f"Chunk {i + 1} ({content_type}):\n"
preview += f" Length: {len(chunk.page_content)} characters\n"
preview += f" Has Table: {has_table}, Has Code/Environment: {has_code}\n"
preview += f" Metadata: {chunk.metadata}\n"
preview += f" Content preview: {chunk.page_content[:100]}...\n"
preview += "-" * 30 + "\n"
if len(chunks) > max_chunks:
preview += f"... and {len(chunks) - max_chunks} more chunks\n"
return preview
# Global unified chunker instance that supports both Markdown and LaTeX
document_chunker = UnifiedDocumentChunker() |