theano-BPR

============ Using Theano to implement the matrix factorization with BPR ranking loss, as described in:

Steffen Rendle, et al. BPR: Bayesian personalized ranking from implicit feedback UAI'09

Optimizer: Batched SGD

Usage:

Required softwares: numpy, theano
$ python run_example.py

Evaluation:

TopK recommendation (hold-1-out evaluation), with measures Hit Ratio and NDCG. More details about the evaluation can be found in:

Xiangnan He, et al. Fast Matrix Factorization for Online Recommendation with Implicit Feedback SIGIR'16

Welcome any comments for improving the efficency!

Author:

Xiangnan He (http://www.comp.nus.edu.sg/~xiangnan/)

Contact: xiangnanhe@gmail.com