File size: 11,667 Bytes
d520909 |
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 |
"""
Semantic Chunking Utilities
Strategies for splitting and merging document content
into semantically meaningful chunks.
"""
import re
from dataclasses import dataclass
from typing import Any, Dict, List, Optional, Tuple
from ..chunks.models import (
BoundingBox,
ChunkType,
DocumentChunk,
)
@dataclass
class ChunkingConfig:
"""Configuration for semantic chunking."""
# Size limits
min_chunk_chars: int = 50
max_chunk_chars: int = 2000
target_chunk_chars: int = 500
# Overlap for context preservation
overlap_chars: int = 100
# Splitting behavior
split_on_headings: bool = True
split_on_paragraphs: bool = True
preserve_sentences: bool = True
# Merging behavior
merge_small_chunks: bool = True
merge_threshold_chars: int = 100
class SemanticChunker:
"""
Semantic chunking engine.
Splits text into meaningful chunks based on document structure,
headings, paragraphs, and sentence boundaries.
"""
# Patterns for text splitting
HEADING_PATTERN = re.compile(r'^(?:#{1,6}\s+|[A-Z0-9][\.\)]\s+|\d+[\.\)]\s+)', re.MULTILINE)
PARAGRAPH_PATTERN = re.compile(r'\n\s*\n')
SENTENCE_PATTERN = re.compile(r'(?<=[.!?])\s+(?=[A-Z])')
def __init__(self, config: Optional[ChunkingConfig] = None):
self.config = config or ChunkingConfig()
def chunk_text(
self,
text: str,
metadata: Optional[Dict[str, Any]] = None,
) -> List[Dict[str, Any]]:
"""
Split text into semantic chunks.
Args:
text: Input text to chunk
metadata: Optional metadata to include with each chunk
Returns:
List of chunk dictionaries with text and metadata
"""
if not text or not text.strip():
return []
metadata = metadata or {}
chunks: List[Dict[str, Any]] = []
# Split by headings first
if self.config.split_on_headings:
sections = self._split_by_headings(text)
else:
sections = [{"heading": None, "text": text}]
for section in sections:
section_chunks = self._chunk_section(
section["text"],
section.get("heading"),
)
for chunk_text in section_chunks:
if len(chunk_text.strip()) >= self.config.min_chunk_chars:
chunks.append({
"text": chunk_text.strip(),
"heading": section.get("heading"),
**metadata,
})
# Merge small chunks
if self.config.merge_small_chunks:
chunks = self._merge_small_chunks(chunks)
return chunks
def _split_by_headings(self, text: str) -> List[Dict[str, Any]]:
"""Split text by heading patterns."""
sections = []
current_heading = None
current_text = []
lines = text.split("\n")
for line in lines:
if self.HEADING_PATTERN.match(line):
# Save previous section
if current_text:
sections.append({
"heading": current_heading,
"text": "\n".join(current_text),
})
current_heading = line.strip()
current_text = []
else:
current_text.append(line)
# Save last section
if current_text:
sections.append({
"heading": current_heading,
"text": "\n".join(current_text),
})
return sections if sections else [{"heading": None, "text": text}]
def _chunk_section(
self,
text: str,
heading: Optional[str],
) -> List[str]:
"""Chunk a single section."""
if len(text) <= self.config.max_chunk_chars:
return [text]
# Split by paragraphs
if self.config.split_on_paragraphs:
paragraphs = self.PARAGRAPH_PATTERN.split(text)
else:
paragraphs = [text]
chunks = []
current_chunk = ""
for para in paragraphs:
para = para.strip()
if not para:
continue
# Check if adding this paragraph exceeds limit
if len(current_chunk) + len(para) + 1 <= self.config.target_chunk_chars:
if current_chunk:
current_chunk += "\n\n" + para
else:
current_chunk = para
else:
# Save current and start new
if current_chunk:
chunks.append(current_chunk)
# If paragraph is too long, split further
if len(para) > self.config.max_chunk_chars:
sub_chunks = self._split_long_text(para)
chunks.extend(sub_chunks[:-1])
current_chunk = sub_chunks[-1] if sub_chunks else ""
else:
current_chunk = para
if current_chunk:
chunks.append(current_chunk)
return chunks
def _split_long_text(self, text: str) -> List[str]:
"""Split long text by sentences."""
if not self.config.preserve_sentences:
# Simple character-based split
return self._split_by_chars(text)
sentences = self.SENTENCE_PATTERN.split(text)
chunks = []
current_chunk = ""
for sentence in sentences:
sentence = sentence.strip()
if not sentence:
continue
if len(current_chunk) + len(sentence) + 1 <= self.config.target_chunk_chars:
if current_chunk:
current_chunk += " " + sentence
else:
current_chunk = sentence
else:
if current_chunk:
chunks.append(current_chunk)
if len(sentence) > self.config.max_chunk_chars:
# Sentence too long - split by chars
sub_chunks = self._split_by_chars(sentence)
chunks.extend(sub_chunks[:-1])
current_chunk = sub_chunks[-1] if sub_chunks else ""
else:
current_chunk = sentence
if current_chunk:
chunks.append(current_chunk)
return chunks
def _split_by_chars(self, text: str) -> List[str]:
"""Split text by character count with overlap."""
chunks = []
start = 0
text_len = len(text)
while start < text_len:
end = min(start + self.config.target_chunk_chars, text_len)
# Try to break at word boundary
if end < text_len:
# Look for last space before limit
space_idx = text.rfind(" ", start, end)
if space_idx > start:
end = space_idx
chunks.append(text[start:end].strip())
# Apply overlap
start = end - self.config.overlap_chars
if start < 0 or start >= text_len:
break
return chunks
def _merge_small_chunks(
self,
chunks: List[Dict[str, Any]],
) -> List[Dict[str, Any]]:
"""Merge chunks smaller than threshold."""
if not chunks:
return chunks
merged = []
current = None
for chunk in chunks:
text = chunk["text"]
if current is None:
current = chunk.copy()
continue
# Check if should merge
current_len = len(current["text"])
new_len = len(text)
if (current_len < self.config.merge_threshold_chars and
current_len + new_len <= self.config.max_chunk_chars and
current.get("heading") == chunk.get("heading")):
# Merge
current["text"] = current["text"] + "\n\n" + text
else:
merged.append(current)
current = chunk.copy()
if current:
merged.append(current)
return merged
class DocumentChunkBuilder:
"""
Builder for creating DocumentChunk objects.
Provides a fluent interface for chunk construction with
automatic ID generation and validation.
"""
def __init__(
self,
doc_id: str,
page: int,
):
self.doc_id = doc_id
self.page = page
self._chunks: List[DocumentChunk] = []
self._sequence_index = 0
def add_chunk(
self,
text: str,
chunk_type: ChunkType,
bbox: BoundingBox,
confidence: float = 1.0,
metadata: Optional[Dict[str, Any]] = None,
) -> "DocumentChunkBuilder":
"""Add a chunk."""
chunk_id = DocumentChunk.generate_chunk_id(
doc_id=self.doc_id,
page=self.page,
bbox=bbox,
chunk_type_str=chunk_type.value,
)
chunk = DocumentChunk(
chunk_id=chunk_id,
doc_id=self.doc_id,
chunk_type=chunk_type,
text=text,
page=self.page,
bbox=bbox,
confidence=confidence,
sequence_index=self._sequence_index,
metadata=metadata or {},
)
self._chunks.append(chunk)
self._sequence_index += 1
return self
def add_text(
self,
text: str,
bbox: BoundingBox,
confidence: float = 1.0,
) -> "DocumentChunkBuilder":
"""Add a text chunk."""
return self.add_chunk(text, ChunkType.TEXT, bbox, confidence)
def add_title(
self,
text: str,
bbox: BoundingBox,
confidence: float = 1.0,
) -> "DocumentChunkBuilder":
"""Add a title chunk."""
return self.add_chunk(text, ChunkType.TITLE, bbox, confidence)
def add_heading(
self,
text: str,
bbox: BoundingBox,
confidence: float = 1.0,
) -> "DocumentChunkBuilder":
"""Add a heading chunk."""
return self.add_chunk(text, ChunkType.HEADING, bbox, confidence)
def add_paragraph(
self,
text: str,
bbox: BoundingBox,
confidence: float = 1.0,
) -> "DocumentChunkBuilder":
"""Add a paragraph chunk."""
return self.add_chunk(text, ChunkType.PARAGRAPH, bbox, confidence)
def build(self) -> List[DocumentChunk]:
"""Build and return the list of chunks."""
return self._chunks.copy()
def reset(self) -> "DocumentChunkBuilder":
"""Reset the builder."""
self._chunks = []
self._sequence_index = 0
return self
def estimate_tokens(text: str) -> int:
"""
Estimate token count for text.
Uses simple heuristic: ~4 characters per token.
"""
return len(text) // 4
def split_for_embedding(
text: str,
max_tokens: int = 512,
overlap_tokens: int = 50,
) -> List[str]:
"""
Split text for embedding model input.
Args:
text: Text to split
max_tokens: Maximum tokens per chunk
overlap_tokens: Overlap between chunks
Returns:
List of text chunks
"""
max_chars = max_tokens * 4
overlap_chars = overlap_tokens * 4
config = ChunkingConfig(
max_chunk_chars=max_chars,
target_chunk_chars=max_chars - 100,
overlap_chars=overlap_chars,
)
chunker = SemanticChunker(config)
chunks = chunker.chunk_text(text)
return [c["text"] for c in chunks]
|