scipreprocess 0.1.0

A modular pipeline for preprocessing scientific documents (PDF, DOCX, TEX, XML, TXT)

SciPreprocess

A modular, open-source pipeline for preprocessing scientific documents in multiple formats (PDF, DOCX, LaTeX, JATS XML, TXT) for LLM consumption and NLP analysis.

Features

• 📄 Multi-format support: PDF, DOCX, TEX, JATS XML, and plain text
• 🔍 OCR support: Extract text from scanned documents with Tesseract
• 🧹 Text cleaning: Remove citations, normalize unicode, clean special characters
• 🔤 NLP processing: Tokenization, lemmatization, stopword removal using spaCy or NLTK
• 📑 Section detection: Automatically identify paper sections (Abstract, Introduction, etc.)
• 🔗 Acronym handling: Detect and expand acronyms using scispacy
• 📊 Feature extraction: TF-IDF and semantic embeddings with sentence-transformers
• 🔎 Semantic search: FAISS indexing for efficient similarity search
• 🧩 Modular design: Use only the components you need

Installation

From PyPI (Recommended)

pip install scipreprocess

With Optional Dependencies

Install specific feature sets:

# PDF support
pip install "scipreprocess[pdf]"

# NLP features
pip install "scipreprocess[nlp]"

# Machine learning features
pip install "scipreprocess[ml]"

# OCR support
pip install "scipreprocess[ocr]"

# Everything
pip install "scipreprocess[all]"

Development Installation

For development or from source:

git clone https://github.com/Tarikul-Islam-Anik/scipreprocess.git
cd scipreprocess
pip install -e ".[all,dev]"

Post-Installation Setup

For NLP features, download required models:

# Download spaCy model
python -m spacy download en_core_web_sm

# Install scispacy model (optional but recommended)
pip install https://s3-us-west-2.amazonaws.com/ai2-s2-scispacy/releases/v0.5.1/en_core_sci_sm-0.5.1.tar.gz

Quick Start

Basic Usage

from scipreprocess import preprocess_file

# Process a single document
doc_json, clean_text = preprocess_file("path/to/paper.pdf")

# Access the results
print(doc_json['metadata']['title'])
print(doc_json['abstract'])
print(doc_json['sections'])
print(doc_json['acronyms'])

Process Multiple Documents

from scipreprocess import preprocess_documents

# Process multiple documents
files = ["paper1.pdf", "paper2.docx", "paper3.tex"]
results = preprocess_documents(files)

# Access results
documents = results['documents']
tfidf_matrix = results['tfidf']['X']
vectorizer = results['tfidf']['vectorizer']
chunks = results['chunks']
embeddings = results['embeddings']  # if enabled

Custom Configuration

from scipreprocess import PipelineConfig
from scipreprocess.pipeline import PreprocessingPipeline

# Configure the pipeline
config = PipelineConfig(
    use_ocr=True,
    use_spacy=True,
    use_semantic_embeddings=True,
    spacy_model='en_core_sci_sm',
    embedding_model='sentence-transformers/all-MiniLM-L6-v2',
    chunk_target_sentences=(3, 8)
)

# Create pipeline with custom config
pipeline = PreprocessingPipeline(config)
doc_json, text = pipeline.preprocess_file("paper.pdf")

Pipeline Components

The pipeline is organized into modular components:

• parsers: Document ingestion (PDF, DOCX, TEX, XML, TXT)
• preprocessing: Text cleaning, tokenization, lemmatization
• acronyms: Acronym detection and expansion
• sectioning: Section splitting and chunking
• features: TF-IDF and semantic embeddings
• pipeline: Main orchestration

Architecture

scipreprocess/
├── config.py          # Configuration dataclasses
├── models.py          # Data models (ParsedDocument)
├── utils.py           # Dependency management and helpers
├── parsers.py         # Document parsers for each format
├── preprocessing.py   # Text cleaning and NLP
├── acronyms.py        # Acronym detection/expansion
├── sectioning.py      # Section splitting and chunking
├── features.py        # Feature extraction (TF-IDF, embeddings)
└── pipeline.py        # Main pipeline orchestration

Output Format

The pipeline produces structured JSON for each document:

{
    "metadata": {
        "title": "Paper Title",
        "source_file": "path/to/file.pdf",
        "pages": 12
    },
    "abstract": "Abstract text...",
    "sections": [
        {"heading": "Introduction", "text": "..."},
        {"heading": "Methods", "text": "..."},
        ...
    ],
    "acronyms": {
        "NLP": "Natural Language Processing",
        "ML": "Machine Learning"
    },
    "figures": [],
    "tables": [],
    "equations": [],
    "references": []
}

Dependencies

Required

• unidecode: Unicode normalization

Optional

• PyMuPDF: PDF parsing
• python-docx: DOCX parsing
• lxml: XML parsing
• opencv-python + pytesseract: OCR support
• nltk: Basic NLP (tokenization, stopwords, lemmatization)
• spacy + scispacy: Advanced NLP and abbreviation detection
• pysbd: Sentence boundary detection
• scikit-learn: TF-IDF vectorization
• sentence-transformers: Semantic embeddings
• faiss: Similarity search

Development

Setup Development Environment

# Clone the repository
git clone https://github.com/Tarikul-Islam-Anik/scipreprocess.git
cd scipreprocess

# Install in development mode with dev dependencies
pip install -e ".[all,dev]"

# Run tests
pytest

# Format code
black src/ tests/

# Lint code
ruff check src/ tests/

# Type checking
mypy src/

Documentation

• Examples: examples/basic_usage.py

Running Tests

# Run all tests
pytest

# Run with coverage
pytest --cov=scipreprocess --cov-report=html

Contributing

Contributions are welcome! Please feel free to submit a Pull Request.

1. Fork the repository
2. Create your feature branch (git checkout -b feature/amazing-feature)
3. Commit your changes (git commit -m 'Add some amazing feature')
4. Push to the branch (git push origin feature/amazing-feature)
5. Open a Pull Request

License

This project is licensed under the MIT License - see the LICENSE file for details.

Citation

If you use this pipeline in your research, please cite:

@software{scipreprocess,
  title = {SciPreprocess: A Modular Scientific Document Preprocessing Pipeline},
  author = {Anik, Tarikul Islam},
  year = {2025},
  url = {https://github.com/Tarikul-Islam-Anik/scipreprocess}
}

Acknowledgments

• Built with spaCy, scispacy, and sentence-transformers
• Inspired by the needs of scientific text processing and NLP research