Cleora

Cleora is an open-source graph embedding library with a Rust core and a Python package, pycleora. It produces node embeddings for recommendation, fraud and social graph work without random-walk sampling or GPUs.

Instead of sampling walks and training a skip-gram model, it builds a sparse transition matrix and multiplies embeddings by it a few times, averaging and normalizing each node's neighbours. Three or four iterations capture co-occurrence similarity and seven or more give context-style similarity. The same input always gives the same output.

It natively handles heterogeneous graphs, bipartite graphs and hypergraphs from TSV files with typed columns, so little preprocessing is needed. The install is about 5 MB, and the site documents eight algorithms, sampling, tuning, compression and benchmarks across five datasets.