Multi-Graph Convolutional Neural Networks
Switch branches/tags
Nothing to show
Clone or download
Latest commit 7584341 Apr 13, 2018
Failed to load latest commit information.
Data Repository initialization. Oct 7, 2017
Notebooks Repository initialization. Oct 7, 2017
pic Modified pictures Oct 7, 2017
LICENSE Initial commit Oct 7, 2017 Update Apr 13, 2018

Multi-Graph Convolutional Neural Networks

The code contained in this repository represents a TensorFlow implementation of the Recurrent Muli-Graph Convolutional Neural Network depicted in:

Geometric Matrix Completion with Recurrent Multi-Graph Neural Networks (in Proc. NIPS, 2017)
Federico Monti, Michael M. Bronstein, Xavier Bresson

Repository Structure

The repository is organized in two main folders: Notebooks and Data. Notebooks contains the python scripts we exploited for realizing the experiments depicted in the paper. Data, the datasets we used.

Every single folder is divided in 5 different subfolders, representing the 5 different datasets we used in our experiments: Movielens 100K, Douban, Flixster, Yahoo Music and our Synthetic Dataset.

When shall I use MGCNN?

MGCNN is a Multi-Graph CNN able to operate on signals defined over multiple graphs. In the paper we exploited this solution for solving the recommendation problem. However, the architecture is general and can be used for any multi-graph dimensional signal.

Useful links