cleora

Why we included this project

Cleora turns graph-structured data into dense entity vectors without needing a GPU, negative sampling, or any of the usual deep-learning plumbing. It computes all possible random walks in a single matrix multiplication, which keeps it fast and deterministic enough to retrain on a schedule in production. That makes it a practical fit for recommendation, fraud detection, social network analysis, and knowledge graph work, where you feed in user-item interactions or entity triples and get embeddings for clustering, link prediction, or similarity search. It installs with pip, weighs in around 5 MB, and the Synerise team behind it has placed in several industry recommendation challenges, so the approach has been tested on real data rather than just benchmarks. If you want embeddings from relational data without standing up a deep-learning stack, this is a lightweight option that gets you there quickly.