neural_collaborative_filtering

Why we included this project

The authors' own implementation of the Neural Collaborative Filtering paper, this codebase ships the three models from that WWW'17 work: Generalized Matrix Factorization, a multi-layer perceptron version, and NeuMF, which combines the two. If you want to understand how neural networks get past the dot-product scoring of classic matrix factorization, this is a clear place to start. The training pipeline targets implicit feedback and ranking, using log loss with negative sampling, so it is well suited to reproducing the paper's results or extending the approach. It is research code, though, so expect to adapt it before anything goes into production.