KBGNN

This is the pytorch implementation for our ICDM 2022 paper:

Wei Ju, Yifang Qin, Ziyue Qiao, Xiao Luo, Yifan Wang, Yanjie Fu, and Ming Zhang(2022). Kernel-based Substructure Exploration for Next POI Recommendation

In this paper, we propose a Kernel-Based Graph Neural Network (KBGNN) for next POI recommendation, which combines the characteristics of both geographical and sequential influences in a collaborative way.

Please cite our paper if you use the code.

Environment Requirement

The code has been tested running under Python 3.8.13. The required packages are as follows:

• pytorch == 1.11.0
• torch_geometric == 2.0.4
• pandas == 1.4.1
• sklearn == 0.23.2 ``

Running Example

For example, to generate Foursquare-Tokyo data for KBGNN models, firstly the raw data should be downloaded and unzipped at ~/raw_data/.

After the download, run:

mkdir processed && cd processed
mkdir tky
cd ../utils
python process_data.py

which will generate processed data files under the directory ~/processed/tky/.

To conduct experiment on Foursquare-Tokyo, run:

cd ./model
python main.py --data tky --batch 1024 --patience 10 --gcn_num 2 --max_step 2

For more execution arguments of KBGNN, please refer to ~/model/main.py or run

python main.py -h