dawenl/cofactor

Implements weighted matrix factorization regularized by item co-occurrence embeddings to improve collaborative filtering recommendations, leveraging item similarity structure beyond user-item interactions. Built on NumPy/SciPy with scikit-learn dependencies, the approach combines traditional matrix factorization with learned item embeddings that capture co-occurrence patterns. Includes preprocessing pipelines and benchmarks on Taste Profile and MovieLens-20M datasets.

166 stars. No commits in the last 6 months.