Code for CS6208 project

Music Recommendation with Graph ConvNet

Ou Longshen's project code submission.

Task: edge regression

Dataset

Last.fm -2k dataset: https://grouplens.org/datasets/hetrec-2011/

• 1,892 users
• 17,632 artists
• 92,834 pairs of user-artist playing event count (Number of edges is too large for full batch training. Stochastic subgraph sampling is adopted.)

Model

Graph Convolutional Networks (GCN) https://arxiv.org/abs/1609.02907

The implementation of graph convolutional layers in the code was inspired by and adapted from the Deep Graph Library (DGL): https://docs.dgl.ai/_modules/dgl/nn/pytorch/conv/graphconv.html#GraphConv

Environment

# On linux (CUDA 11.7)
pip install torch==1.13.1+cu117 torchvision==0.14.1+cu117 torchaudio==0.13.1 --extra-index-url https://download.pytorch.org/whl/cu117
pip install  dgl -f https://data.dgl.ai/wheels/cu117/repo.html
pip install  dglgo -f https://data.dgl.ai/wheels-test/repo.html

Files

data            Dataset used in the experiment.
data_analysis.py/ipynb  Data analysis and preparation
models_gcn.py   GCN implementation and improvement
dataset.py      Dataset class    
train.py        Code for model training and evaluation.
train.ipynb     Training log.
utils.py        Util class and functions

Run the code

Check the train.ipynb for previous training logs.

Or use below command:

python train.py

to run the training script.