This is code for our WWWJ paper《HMGCL: Heterogeneous Multigraph Contrastive Learning for LBSN Friend Recommendation》
Before to execute HMGCL, it is necessary to install the following packages:
pip install dgl
pip install torch
pip install scikit-learn
- numpy ==1.13.1
- torch ==1.7.1
- scikit-learn==1.0.2
- dgl ==0.7.2
You can download whole raw Foursquare Dataset here.
Our data can be found at here.
- --run main.py to train the HMGCL. and it probably need at least 11G GPU memory
- --run test.py to estimate the performance of HMGCL 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.
Note: This is only a reference implementation of HMGCL. Our code implementation is partially based on the DGL library, for which we are grateful.