Code and instructions to reproduce the analysis in the paper.

You can find online videos that describe IMINFECTOR and its previous variant.

mkdir Code Data Figures
cd Code
git clone https://github.com/GiorgosPanagopoulos/Influence-Maximization-via-Representation-Learning

To run this code you will need the following python packages:

• numpy
• pandas
• scipy
• tensorflow-gpu
• igraph
• pyunpack
• patool

which can be installed using the requirements.txt:

pip install -r requirements.txt

All datasets need certain preprocessing before the experiments.

python preprocessing

The script creates the required folder structure for every dataset (Digg, Weibo, MAG)->Init_Data,Embeddings, Seeds, Spreading. It then downloads the Digg and Weibo datasets, and preprocesses them for curation and derivation of the network and the diffusion cascades.
To derive the MAG network and diffusion cascades, we employed the tables Paper, Paper References, Author, PaperAuthorAffiliation, Fields of Study, Paper Fields of Study from the official MAG. There is also an open version. Add these datasets to "MAG/Init_data" and run mag_preprocessing.py

The main function will derive and evaluate the seed sets of the two metrics and IMINFECTOR as well as the input for the baseline methods.
However, some of the baselines need to be run separately using their original codes, found in these locations:
Credit Distribution and Simpath
IMM

Run with default parameters for sampling percentage, learning rate, number of epochs, embeddings size and number of negative samples.

python main --sampling_perc=120 learning_rate=0.1 --n_epochs=5 --embedding_size=50 --num_neg_samples=10

Manually change the three paths to the directories of the datasets in plot_precision.R and plot_spreading.R and run them.

If you use this work, please cite:

@article{panagopoulos2020multi,
  title={Multi-task Learning for Influence Estimation and Maximization},
  author={Panagopoulos, George and Malliaros, Fragkiskos and Vazirgiannis, Michalis},
  journal={IEEE Transactions on Knowledge and Data Engineering},
  year={2020},
  publisher={IEEE}
}

@inproceedings{panagopoulos2020influence,
  title={Influence Maximization Using Influence and Susceptibility Embeddings},
  author={Panagopoulos, George and Malliaros, Fragkiskos D and Vazirgianis, Michalis},
  booktitle={Proceedings of the International AAAI Conference on Web and Social Media},
  volume={14},
  pages={511--521},
  year={2020}
}

License

• MIT License