Paper | Poster

Official code implementation of AAAI 2025 paper "Self-Explainable Graph Transformers for Link Sign Prediction".This work presents SE-SGformer, a self-explainable framework for signed graphs. It introduces a new Transformer architecture with positional encoding based on signed random walks, offering superior expressive power over existing methods. SE-SGformer replaces neural network-based decoders with a novel decision process that uses $K$-nearest (farthest) neighbors to predict edge signs, providing critical interpretability in decision-making.The overall architecture of SE-SGformer is illustrated as follows:

pip install -r requirements.txt

• Python 3.9.12
• torch = 2.3.1
• numpy~=1.26.4
• scikit-learn~=1.5.1
• pandas~=2.2.2
• tqdm~=4.64.0
• scipy~=1.10.1
• torch_geometric~=2.5.3
• torchvision~=0.18.1
• torchaudio~=2.3.1

We divided the dataset five times, with each division splitting the training and test sets at an 8:2 ratio. Five experiments were conducted, and the average of the results was taken.

To run the encoder module:

python main.py 

To run the decoder module:

python -u discriminator.py --dataset Epinions --offset 4

dataset:Epinions, amazon-music, KuaiRand, KuaiRec,soc-sign-bitcoinotc,soc-sign-bitcoinalpha,WikiElec,WikiRfa offset: for Epinions and amazon-music,offset=4,for KuaiRand and KuaiRec , offset = 1 , for others , offset = 0. Here, "offset" refers to the threshold used to distinguish between positive and negative edges.

python -u discriminator.py --dataset Epinions --offset 4

python -u discriminator.py --dataset amazon-music --offset 4

python -u discriminator.py --dataset KuaiRand --offset 1

python -u discriminator.py --dataset KuaiRec --offset 1

python -u discriminator.py --dataset BitcoinOTC --offset 0

python -u discriminator.py --dataset Bitcoin-alpha --offset 0

python -u discriminator.py --dataset WikiElec --offset 0

python -u discriminator.py --dataset WikiRfa --offset 0