Scalable Continuous-time Diffusion Framework for Network Inference and Influence Estimation

FIM is framework to approximate the continuous-time diffusion from cascade data. It is established by utilizing the continuous-time dynamical system. In this paper, FIM is proposed for network inference from cascades and influence estimation on the inferred networks.

This repository contains the source codes of FIM. For further details, please refer to our paper in WWW 2024 (https://arxiv.org/abs/2403.02867). Should you encounter any issues, please reach out to Keke Huang, thanks!

Requirements

• CUDA 10.1.243
• python 3.6.10
• pytorch 1.4.0
• GCC 5.4.0
• cnpy
• swig-4.0.1

Compilation

make

Command example

python train.py --seed 25190 --dataset HR --steps 2500 --patience 1000 --lr 0.001 --l1_lambda 0.01 --batch 50 --tau 0.5 --norm 0.5  --train_per 0.8 --val_per 0.1

Citation

Please cite our paper if it is relevant to your work, thanks!

@inproceedings{HuangGCX24,
  author       = {Keke Huang and
                  Ruize Gao and
                  Bogdan Cautis and
                  Xiaokui Xiao},
  title        = {Scalable Continuous-time Diffusion Framework for Network Inference
                  and Influence Estimation},
  booktitle    = {{WWW}},
  pages        = {2660--2671},
  year         = {2024},
}