Official implementation of our CIKM 2025 work "Structure-prior Informed Diffusion Model for Graph Source Localization with Limited Data"
run command:
python main_newdiff.py --dataset [dataset]
Available datasets:(as in datasets\synthesis)
-
android_10_30, christianity_10_30, jazz_IC50, jazz_LT50, jazz_SIS50, netscience_IC50, netscience_LT50, netscience_SIS50, power_IC50, power_LT50, power_SIS50,
-
Some large datasets files are not included in the repository. Find them in the following link: https://drive.google.com/drive/folders/1Ga7qTNscLug49o69E18AAyiXvRKDjjY5?usp=sharing
Tunable params can be checked in the following description.
run command:
python main_newdiff.py --dataset [dataset] --save_dict [path to saved model]
e.g.
python main_newdiff.py --dataset jazz_IC50 --state_dict ./saved_diffusers/jazz_IC50_best.pt
Please check requirements.txt for the required packages.
dataset
: Name of the dataset to usegnn_type
: Type of GNN architecture (default: gcn)noise_emb_dim
: Dimension of noise embedding (default: 128)hidden_dim
: Hidden layer dimension (default: 128)num_layers
: Number of GNN layers (default: 5)activation
: Activation function (default: prelu)mlp_layers
: Number of MLP layers (default: 4)num_advisors
: Number of advisor models (default: 1)
feat_drop
: Feature dropout rate (default: 0.0)attn_drop
: Attention dropout rate (default: 0.0)negative_slope
: Negative slope for LeakyReLU (default: 0.2)residual
: Whether to use residual connections (default: True)scheduler
: Whether to use learning rate scheduler (default: True)train_cond
: Whether to train the condition module (default: False)
@article{chen2025structure,
title={Structure-prior Informed Diffusion Model for Graph Source Localization with Limited Data},
author={Chen, Hongyi and Ding, Jingtao and Liang, Xiaojun and Li, Yong and Zhang, Xiao-Ping},
journal={arXiv e-prints},
pages={arXiv--2502},
year={2025}
}