GraphUIL

Our Pytorch implementation of Graph Neural Networks for User Identity Linkage.

1. Requirements

To install requirements:

pip install -r requirements.txt

2. Repository Structure

• data/: contains the processed data.

  • graph/: adj_s.pkl, adj_t.pkl: adjacency matrices of the source network and the target network. embeds.pkl: textual input features of two networks.
  • label/: anchor files, train_test_0.x.pkl splits the training anchors at ratios range from 0.1 to 0.9.

  The dataset Douban-Weibo is provided by the PHD student Siyuan Chen. If you use the data, please cite the following paper. More details refer to INFUNE.

  @inproceedings{chen2020infune,
     title={A Novel Framework with Information Fusion and Neighborhood Enhancement for User Identity Linkage},
     author={Chen, Siyuan and Wang, Jiahai and Du, Xin and Hu, Yanqing},
     booktitle={24th European Conference on Artificial Intelligence (ECAI)},
     pages={1754--1761},
     year={2020}
  }
  

• logs/: saving logs

• models/: contains loss function and metric for evaluation.

  • base.py
  • loss.py
  • netEncode.py: GNN layer.

• UIL/GraphUIL.py: GraphUIL framework.

• utils/: tool functions for processing data and logging.

• config.py: hyperparameters.

• main.py.

3. Runing

python main.py