rezolva

Entity resolution for everyone. Minimal. No dependencies.

rezolva is a lightweight, flexible, and extensible entity resolution library implemented in pure Python. It's designed for simplicity, educational purposes, and easy integration into larger projects. (it is also very much a work in progress)

Features

• Zero dependencies: built with Python standard library only
• Modular architecture: easy to customize and extend
• Defaults: out-of-the-box implementations of ER algorithms
• Flexible: adaptable to various ER scenarios
• Evaluation: utilities for testing precision, recall, F1, accuracy
• Visualization: trace candidates through the resolution process

Installation

pip install rezolva

Quick Start

Here's a simple example of how to use rezolva:

from rezolva import Entity, EntityResolver, SimpleBlocker, SimpleModelBuilder, SimplePreprocessor
from rezolva.matchers import CosineSimilarityMatcher
from rezolva.preprocessors.preprocessing_functions import lowercase, strip_whitespace, remove_punctuation

# Set up components
preprocessor = SimplePreprocessor([lowercase, strip_whitespace, remove_punctuation])
model_builder = SimpleModelBuilder(['title', 'description', 'brand'])
matcher = CosineSimilarityMatcher(threshold=0.5, attribute_weights={'title': 2.0, 'description': 1.5, 'brand': 1.0})
blocker = SimpleBlocker(lambda e: e.attributes['brand'].lower())

# Create resolver
resolver = EntityResolver(preprocessor, model_builder, matcher, blocker)

# Train the resolver
training_entities = [
    Entity("1", {"title": "iPhone 12", "description": "Latest Apple smartphone", "brand": "Apple"}),
    Entity("2", {"title": "Galaxy S21", "description": "Samsung's flagship phone", "brand": "Samsung"}),
    Entity("3", {"title": "Pixel 5", "description": "Google's latest smartphone", "brand": "Google"}),
]
resolver.train(training_entities)

# Resolve new entities
new_entities = [
    Entity("4", {"title": "iPhone 12 Pro", "description": "Apple's premium smartphone", "brand": "Apple"}),
    Entity("5", {"title": "Galaxy S21+", "description": "Samsung's large screen flagship", "brand": "Samsung"}),
]
results = resolver.resolve(new_entities, top_k=1)

# Print results
for entity, matches in results:
    print(f"Top matches for {entity.id} - {entity.attributes['title']}:")
    for match, score in matches[:2]:
        print(f"  Match: {match.id} - {match.attributes['title']} (Score: {score:.2f})")

Advanced Usage

For more advanced examples, including custom preprocessing, different matching algorithms, model saving/loading, deduplication, clustering, and more, please check the examples directory in the repository.

Customization

You can extend rezolva by creating custom implementations of its components:

1. Subclass the base component classes (Preprocessor, ModelBuilder, Matcher, Blocker)
2. Implement the required methods with your custom logic
3. Use your custom components when creating the EntityResolver

Contributing

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

License

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