scpei / neural_bayesian_personalized_ranking

Representation Learning and Pairwise Ranking for Implicit Feedback in Top-N Item Recommendation

Bayesian Personalized Ranking Neural Network

This repository provides a tensorflow implementation of Neural Bayesian Personalized Ranking from implicit feedback for top-N item recommendation.

Pre-Requisite

• Python 2.7
• Numpy
• Tensorflow 1.3

Basic Usage

Example

To run Neural Bayesian Personalized Ranking, execute the following command from the project home directory:
python neural_bpr_v1.py 32 128 0.001 0.01 50

Current neural structure is input layer -> embedding layer -> one hidden layer with relu activation function -> output layer with BPR loss

To run the Bayesian Personalized Ranking under Matrix Factorization model, execute the following commend from the project home directory:
python bpr_loss_mf.py 32 0.001 0.01 50

Options

You can check out the hyper-parameter options using:
python neural_bpr_v1.py --help

Dataset

Benchmark MovieLens 1M Dataset (https://grouplens.org/datasets/movielens/)

Output

The output is pairwise ranking loss, HitRate@10, Normalized Discounted Cumulative Gain@10, Area Under Curve (AUC) in each epoch.