coral-media/php-ir

Information Retrieval algorithms (vector space, similarity, clustering)

v0.5.2 2025-12-22 04:34 UTC

Requires

• php: >=8.2
• ext-curl: *
• ext-json: *
• ext-openssl: *

Requires (Dev)

• friendsofphp/php-cs-fixer: ^3.92
• phpstan/phpstan: ^2.1
• phpunit/phpunit: ^11.5
• symfony/var-dumper: ^7.4

MIT

• Rafael Ernesto Espinosa Santiesteban

• search
• recommendation
• clustering
• k-means
• TF-IDF
• information-retrieval
• text-similarity
• cosine-similarity
• vector-space-model

README

A focused PHP library implementing classical Information Retrieval (IR) algorithms based on the Stanford IR book
Introduction to Information Retrieval.

The goal of this project is to provide a correct, deterministic, and explainable IR core suitable for search, clustering, and recommendation systems.

License

MIT

Scope

• Vector space model (dense and sparse vectors)
• Similarity measures (cosine, euclidean)
• Term weighting:
  • Term Frequency (TF)
  • Inverse Document Frequency (IDF)
  • TF-IDF
  • BM25 (probabilistic ranking)
• Clustering (K-Means, K-Means++)
• Deterministic, explainable algorithms

Non-Goals

• NLP pipelines
• Transformers / embeddings
• Semantic vector databases
• Dataset abstractions
• Framework integrations

This library intentionally focuses on classical IR, not modern NLP.

Philosophy

This library favors:

• Mathematical correctness
• Explicit, inspectable abstractions
• Deterministic behavior
• Minimal hidden state
• Native acceleration via Zephir (optional, future-facing)

API Stability

New functionality will be added in a backward-compatible manner.

Releasing

Project versions are tracked using a .version file and Git tags.

To bump the version:

./scripts/bump-version.sh {major|minor|patch}
git commit -am "chore(release): bump version"