Appending adversarial training on multimedia features enhances the performance of multimedia recommender system.

This is our official implementation for the paper:

Jinhui Tang, Xiangnan He, Xiaoyu Du, Fajie Yuan, Qi Tian, and Tat-Seng Chua, Adversarial Training Towards Robust Multimedia Recommender System.

If you use the codes, please cite our paper. Thanks!

Requirements

• Tensorflow 1.7
• numpy, scipy

Quick Start

0. Data

   • f_resnet.npy Deep image features extracted with Resnet. The $i$-th row indicates the $i$-th item feature.
   • pos.txt The training samples used in training process. The numbers $u$ and $i$ in each row indicate an interaction between user $u$ and item $i$.
   • neg.txt The test samples used in testing process. The first number of row $u$ is the only positive sample in test, the following numbers of row $u$ are the negative samples for user $u$.

1. Pretrained VBPR The pretrained VBPR is stored in weights/best-vbpr.npy

2. Traing AMR

   bash run.sh
   

   The training logs are stored in logs

Source Files

Source files are stored in src/.

• main.py. The main entrance of the program.

• solver/*. The solvers managing the training process.

• model/*. The models.

• dataset/*. The data readers.