HMGNN

This is code for our paper《HMGCL: Heterogeneous Multigraph Contrastive Learning for LBSN Friend Recommendation》

Before to execute HMGNN, it is necessary to install the following packages:
pip install dgl
pip install torch
pip install scikit-learn

Requirements

• numpy ==1.13.1
• torch ==1.7.1
• scikit-learn==1.0.2
• dgl ==0.7.2

Data Set

Due to the size limitation of Supplementary Material, we cannot add the processed data here.

At the same time, due to the limitations of the Double-blind Reviewing Process, current links such as Google's drive often carry personal information, so we feel that it is unsuitable to pass data through these drive links.

Therefore, we have decided to only open the code at this stage. If the paper be accepted, we will publish all the data and other information on Github. You can be sure that we'll try our best!

You can download whole raw Foursquare Dataset here.

Basic Usage

• --run main.py to train the HMGNN. and it probably need at least 11G GPU memory
• --run test.py to estimate the performance of HMGNN based on the user representations that we learned during our experiments. You can also use this code to individually test the effects of your own learned representation.

Miscellaneous

Note: This is only a reference implementation of HMGNN. Our code implementation is partially based on the DGL library, for which we are grateful.