PDF Chunker Library v2.0

A production-ready Python library for intelligently chunking PDF documents using sophisticated font analysis, enhanced content filtering, and strategic header detection.

🚀 Features

• Strategic Header Chunking: Advanced font-size analysis with frequency-based header selection
• Enhanced Meaning Detection: AI-powered content analysis with metadata pattern filtering
• Multi-Level Processing: Undersized → Oversized → Hierarchical sub-chunking pipeline
• Robust Content Filtering: Removes document metadata, page markers, and meaningless fragments
• Smart Chunk Processing: Intelligent merging of meaningful short chunks
• Professional Summarization: Extractive summaries with rich metadata output
• Dual Usage Modes: Simple convenience methods AND advanced custom processing
• Multiple Output Formats: JSON, CSV, and custom formats with rich metadata

📦 Installation

Basic Installation

pip install PyMuPDF pypdf

🎯 Quick Start - Two Approaches

🟢 Approach 1: Simple Convenience (Recommended for Most Users)

Perfect for: Quick prototyping, standard use cases, minimal configuration

from pdf_chunker_for_rag.chunk_creator import CleanHybridPDFChunker

# Initialize and process in one line
chunker = CleanHybridPDFChunker()
output_file = chunker.process_and_save('document.pdf')
print(f"✅ Chunks saved to: {output_file}")

Run the example:

cd examples/
python simple_usage.py

What you get:

• Automatic header detection and chunking
• JSON output with metadata
• Multiple format options (JSON/CSV)
• Error handling and validation

🔵 Approach 2: Advanced Custom Processing

Perfect for: Custom applications, data analysis, integration with other systems

from pdf_chunker_for_rag.chunk_creator import CleanHybridPDFChunker

# Get raw chunk data for custom processing
chunker = CleanHybridPDFChunker()
chunks, headers = chunker.strategic_header_chunking('document.pdf')

# Now you have direct access to chunk data
for chunk in chunks:
    topic = chunk['topic']
    content = chunk['content'] 
    word_count = chunk['word_count']
    # Your custom logic here...

# Save however you want
import json
with open('my_chunks.json', 'w') as f:
    json.dump({'chunks': chunks}, f, indent=2)

Your own save logic here

**Run the example:**
```bash
cd examples/
python advanced_usage.py

What you get:

• Direct access to chunk data and headers
• Custom filtering and analysis
• Multiple output formats with custom metadata
• Advanced statistics and reporting

With Enhanced NLP (recommended)

pip install PyMuPDF pypdf spacy
python -m spacy download en_core_web_sm

Development Installation

pip install -e .[dev,nlp]

Quick Start

from pdf_chunker_for_rag import CleanHybridPDFChunker

# Initialize the production chunker
chunker = CleanHybridPDFChunker()

# Process PDF with strategic header chunking
chunks = chunker.strategic_header_chunking(
    pdf_path="your_document.pdf",
    target_words_per_chunk=200
)

print(f"✅ Created {len(chunks)} structured chunks")
print(f"📊 Average chunk size: {sum(c.get('word_count', 0) for c in chunks) // len(chunks)} words")

# Access chunk data
for chunk in chunks:
    print(f"📖 {chunk['topic']} ({chunk['word_count']} words)")
    print(f"📋 {chunk['summary']}")
    print()

Advanced Usage

from pdf_chunker_for_rag import PDFChunker, ChunkingConfig, SummarizationMethod

# Custom configuration
config = ChunkingConfig(
    target_words_per_chunk=300,
    min_header_occurrences=2,
    oversized_threshold=600,
    critical_threshold=1000,
    min_meaningful_words=30,
    summarization_method=SummarizationMethod.EXTRACTIVE
)

chunker = PDFChunker(config)
result = chunker.chunk_pdf("your_document.pdf")

Key Classes

PDFChunker

Main interface for PDF chunking operations.

Methods:

• chunk_pdf(pdf_path): Complete chunking process
• detect_headers(pdf_path): Header detection only
• extract_text(pdf_path): Text extraction only
• get_font_analysis(pdf_path): Font analysis only

ChunkingConfig

Configuration for chunking behavior.

Parameters:

• target_words_per_chunk: Target words per chunk (default: 200)
• min_header_occurrences: Minimum header occurrences for selection (default: 3)
• font_size_tolerance: Tolerance for font size grouping (default: 2.0)
• oversized_threshold: Word count threshold for oversized chunks (default: 500)
• critical_threshold: Critical threshold requiring forced splitting (default: 800)
• min_meaningful_words: Minimum words for meaningful chunks (default: 50)

Data Structures

ChunkData: Represents a processed chunk

• chunk_id: Unique identifier
• topic: Header/topic text
• content: Chunk content
• word_count: Number of words
• summary: Generated summary
• parent_chunk_info: Information about parent chunk (for split chunks)

HeaderData: Represents a detected header

• text: Header text
• font_size: Font size in points
• page: Page number
• is_bold: Whether header is bold

Processing Pipeline

1. Font Analysis: Analyze document fonts and determine normal text size
2. Header Detection: Identify potential headers based on font size
3. Strategic Selection: Select optimal header level using frequency analysis
4. Text Extraction: Extract text with proper reading order
5. Chunk Creation: Create initial chunks based on headers
6. Content Filtering: Remove meaningless content and merge short meaningful chunks
7. Summarization: Generate summaries for all chunks
8. Oversized Processing: Handle large chunks through sub-header detection or forced splitting

Content Quality Features

Meaningless Content Detection

• Version numbers and dates
• Page markers and formatting artifacts
• Low meaningful word ratios
• Incomplete sentences and titles

Smart Merging

• Preserves short but meaningful content
• Forward-direction merging with adjacent chunks
• Maintains topic coherence

NLP-Enhanced Analysis (with spaCy)

• Sentence structure analysis
• Named entity recognition
• Vocabulary diversity scoring
• Professional content detection

Library Architecture

pdf_chunker_for_rag/
├── core/           # Core types and main chunker class
├── analysis/       # Font analysis and header detection
├── filtering/      # Content quality filtering and merging
├── processing/     # Summarization and oversized chunk handling
└── utils/          # Text extraction and utility functions

Examples

Processing Multiple PDFs

import os
from pdf_chunker_for_rag import PDFChunker

chunker = PDFChunker()
results = {}

for filename in os.listdir("pdfs/"):
    if filename.endswith(".pdf"):
        pdf_path = os.path.join("pdfs", filename)
        results[filename] = chunker.chunk_pdf(pdf_path)

# Analyze results
for filename, result in results.items():
    print(f"{filename}: {len(result.chunks)} chunks, "
          f"avg {result.average_chunk_size:.0f} words")

Custom Content Filtering

from pdf_chunker_for_rag.filtering import ContentFilter

# Create custom filter
filter = ContentFilter(min_meaningful_words=30)

# Check if content is meaningful
is_meaningful = filter.has_meaningful_sentence_structure("Your text here")
is_meaningless = filter.is_meaningless_content("Your text here")

Font Analysis Only

from pdf_chunker_for_rag.analysis import FontAnalyzer

analyzer = FontAnalyzer()
font_info = analyzer.analyze_document_fonts("document.pdf")

print(f"Normal text size: {font_info['normal_font_size']:.1f}pt")
print(f"Header threshold: {font_info['min_header_threshold']:.1f}pt")
print(f"Unique font sizes: {len(font_info['all_font_sizes'])}")

Requirements

• Python 3.8+
• PyMuPDF (fitz) >= 1.20.0
• pypdf >= 3.0.0
• spaCy >= 3.4.0 (optional, for enhanced NLP features)

License

MIT License - see LICENSE file for details.

Contributing

1. Fork the repository
2. Create a feature branch
3. Make your changes
4. Add tests if applicable
5. Submit a pull request

Changelog

Version 1.0.0

• Initial release
• Complete modular architecture
• Font-based header detection
• Content quality filtering
• Smart chunk merging
• Multiple summarization methods
• Oversized chunk processing