Fuzzy Context Finder

A Python utility for investigative journalists and researchers that performs intelligent fuzzy keyword searching within documents and extracts customizable context around matched terms. Perfect for text analysis, document investigation, and content exploration where approximate matches and surrounding context are critical.

🔒 Privacy & Security First

All processing happens locally on your machine. No documents, content, or search terms are ever sent to third-party servers or cloud services. This makes Fuzzy Context Finder the ideal choice for investigative journalists who must protect source confidentiality and maintain document security. Whether you're working with leaked documents, confidential sources, or sensitive investigations, your materials never leave your control.

Features

• 🔍 Intelligent Matching Strategies:

  • Lemmatization: Matches word families (coercion → coerced, coercing)
  • Fuzzy Matching: Catches typos (prescription → perscription)
  • Prefix Protection: Prevents false matches (prescription ≠ description)
  • Short Word Family Mode: Matches related words (sex → sexual, sexually)

• 📄 Flexible Context Extraction: Capture customizable amounts of text before, after, or around matched terms

• 🎯 Accuracy-First Design: Built for investigative journalism where missing a match is not an option

• ⚡ Performance Optimized: Caching and efficient algorithms for large documents

Installation

1. Create a Virtual Environment (Recommended)

On Mac/Linux:

python3 -m venv venv
source venv/bin/activate

On Windows:

python -m venv venv
venv\Scripts\activate

2. Install Required Packages

pip install pandas rapidfuzz regex spacy

3. Download spaCy Language Model

python -m spacy download en_core_web_sm

4. Install the Package

From PyPI (when published):

pip install fuzzy-context-finder

From Source:

git clone https://github.com/yourusername/fuzzy_context_finder.git
cd fuzzy_context_finder
pip install -e .

Requirements

Create a requirements.txt file:

pandas>=2.0.0
rapidfuzz>=3.0.0
regex>=2023.0.0
spacy>=3.7.0

Install all at once:

pip install -r requirements.txt
python -m spacy download en_core_web_sm

Quick Start

from fuzzy_context_finder import keyword_context_finder

# Your document content
content = """
The doctor wrote a prescription for antibiotics.
The patient's opioid addiction complicated treatment.
There were allegations of coercion and coerced consent.
"""

# Search terms
search_terms = ["prescription", "addict", "coercion"]

# Find matches with context
results = keyword_context_finder(
    content=content,
    terms=search_terms,
    file_name="medical_report.txt",
    words_around=20,
    match_threshold=80,
    use_lemmatization=True
)

# View results
if results is not None:
    print(results[["Matched Term", "Original Term", "Match Type"]])
    
    # Save to CSV
    results.to_csv("investigation_results.csv", index=False)

Usage Examples

Example 1: Basic Investigation

content = "The investigation revealed sexual misconduct and abuse of power."
search_terms = ["sex", "abuse"]

results = keyword_context_finder(
    content=content,
    terms=search_terms,
    file_name="investigation.txt",
    words_around=15
)

# Matches: "sexual" (from "sex") and "abuse"

Example 2: Multiple Documents

import os

documents = ["report1.txt", "report2.txt", "report3.txt"]
search_terms = ["fraud", "embezzlement", "misconduct"]
all_results = []

for doc_path in documents:
    with open(doc_path, 'r', encoding='utf-8') as f:
        content = f.read()
    
    results = keyword_context_finder(
        content=content,
        terms=search_terms,
        file_name=doc_path,
        match_threshold=85,
        use_lemmatization=True
    )
    
    if results is not None:
        all_results.append(results)

# Combine all results
import pandas as pd
combined_results = pd.concat(all_results, ignore_index=True)
combined_results.to_csv("all_findings.csv", index=False)

Example 3: Adjusting for Precision vs Recall

High Precision (fewer false positives):

results = keyword_context_finder(
    content=content,
    terms=search_terms,
    file_name="document.txt",
    match_threshold=90,  # Stricter threshold
    require_prefix_match=True,  # Require prefix to match
    min_prefix_length=4,  # Longer prefix requirement
    use_lemmatization=True
)

High Recall (catch more variations):

results = keyword_context_finder(
    content=content,
    terms=search_terms,
    file_name="document.txt",
    match_threshold=75,  # Lower threshold
    require_prefix_match=False,  # Allow different prefixes
    use_lemmatization=True,
    word_family_mode=True
)

Parameters

Required Parameters

Parameter	Type	Description
content	str	The text content to search
terms	list	List of search terms (e.g., ["fraud", "corruption"])
file_name	str	Name/identifier for the document

Context Parameters

Parameter	Type	Default	Description
words_before	int	250	Number of words to capture before the matched term
words_after	int	250	Number of words to capture after the matched term
words_around	int	50	Number of words to capture around the term (split evenly)

Matching Strategy Parameters

Parameter	Type	Default	Description
match_threshold	int	80	Minimum similarity score (0-100) for fuzzy matching
require_prefix_match	bool	True	Requires word beginnings to match (prevents "prescription"/"description" confusion)
min_prefix_length	int	3	Number of starting characters that must match exactly
use_lemmatization	bool	True	Use spaCy lemmatization to match word families
word_family_mode	bool	True	Enable prefix matching for short words
family_mode_max_length	int	4	Max term length for family mode (≤ this uses prefix matching)

Match Types Explained

The Match Type column in results shows how each match was found:

• lemma: Matched through lemmatization (e.g., "coercion" → "coerced")
• family: Matched through prefix for short words (e.g., "sex" → "sexual")
• fuzzy: Matched through fuzzy string matching (e.g., typos)

Output Format

Returns a pandas DataFrame with columns:

Column	Description
File Name	Name of the file/document
Page Number	Page number (always 1 for text content)
Matched Term	The actual word found in the text
Original Term	Your search term that matched
Similarity Score	Similarity score (0-100)
Match Type	How the match was found (lemma/family/fuzzy)
Search Term with N Words Context	Context around the term
Previous N Words (Including Term)	Text before the match
Next N Words (Including Term)	Text after the match
Character Position	Position in original text
Word Index	Word position in document

Real-World Use Cases

Investigative Journalism

# Track all mentions of a subject and related terms
search_terms = ["corruption", "bribery", "kickback", "fraud"]
# Lemmatization catches: corrupted, bribes, bribing, fraudulent, etc.

Legal Discovery

# Find all contract-related terms
search_terms = ["contract", "agreement", "terms", "breach"]
# High threshold to avoid false positives in legal documents

Medical Research

# Track medication mentions and variations
search_terms = ["oxycodone", "fentanyl", "prescription", "opioid"]
# Catches: prescriptions, prescribed, opioids, etc.

Troubleshooting

spaCy Model Not Found

If you get an error about missing spaCy model:

python -m spacy download en_core_web_sm

No Matches Found

1. Lower the threshold: Try match_threshold=70
2. Check your search terms: Use root words (e.g., "coerce" instead of "coercion")
3. Disable prefix matching: Set require_prefix_match=False
4. Check lemmatization: Print lemmas to debug:

   import spacy
   nlp = spacy.load("en_core_web_sm")
   doc = nlp("coercion coerced")
   for token in doc:
       print(f"{token.text} → {token.lemma_}")

Too Many False Positives

1. Increase threshold: Try match_threshold=90
2. Enable prefix matching: Set require_prefix_match=True
3. Increase prefix length: Set min_prefix_length=4

Performance Tips

• Large documents: Process in chunks if > 1MB
• Many search terms: Results scale with O(words × terms)
• Caching: Repeated words are cached for efficiency
• Disable unused features: Turn off lemmatization if not needed

Contributing

Contributions welcome! Please:

1. Fork the repository
2. Create a feature branch
3. Add tests for new features
4. Submit a pull request

License

MIT License - see LICENSE file for details

Author

Created for investigative journalists who need accurate, comprehensive document analysis.

Support

• Issues: https://github.com/yourusername/fuzzy_context_finder/issues
• Documentation: https://github.com/yourusername/fuzzy_context_finder/wiki

Changelog

Version 2.0.0

• NEW: Improved acronym handling - now correctly matches A.I., U.S., C.E.O., Ph.D., etc.
• NEW: Acronyms with and without periods now match (searching "AI" finds both "AI" and "A.I.")
• Improved tokenization pattern to preserve acronyms as complete tokens
• Added spaCy lemmatization for better word family matching
• Added prefix matching to prevent false positives
• Added multiple matching strategies (lemma/family/fuzzy)
• Improved tokenization to handle punctuation
• Added caching for performance

Version 1.0.0

• Initial release with basic fuzzy matching