DANI

This is an official implementation for paper titled "DANI: Fast Diffusion Aware Network Inference with Preserving Topological Structure Property".

How to Run C++ Version

The default implementation of DANI is in C++, which is way faster than the python version due to C++ characteristics.

Compile the Code

To compile the code, run the command below.

make

Input Data Format

Only input required for the code is a file containing nodes names and cascades sorted in ascending order regarding infection timestamps. The input file contains two parts which are separated by a blank line:

1. The first $n$ lines of the file contain the name of nodes in the network. Each line contain <id> and <name> where node ids are integers starting from 0 to $n-1$.

   <id>,<name>
   

2. The next $m$ lines contains $m$ cascades, each on a line. Each cascade is represented by a sequence of <id> and the corresponding <timestamp> of infection pairs where <id> and <timestamp> are separated by a comma and pairs are separated by a semi-colon. The cascades are sorted in ascending order regarding infection timestamps.

   <id>,<timestamp>;<id>,<timestamp>;<id>,<timestamp>;...;<id>,<timestamp>
   

An example of input file is in file cascades.txt which contains 10000 cascades generated on the network in dataset/groundtruth/network.txt.

Run the Code

Here is the list of arguments that can be passed to the code:

• -i: which is the address of input file.
• -Y: which is the address of output file.

Run the code using following command:

./main -i:<input_file> -Y:<output_file>

To run the code on the example input file, run the command below:

./main -i:dataset/cascades.txt -Y:dataset/inferred/network.txt

Output

The output of the model is in Tab Separated Triples format. Each line contains a triple in format of $u\space v\space \alpha$ in which alpha is scored assigned to directed edge $(u,\space v)$ by the model. Note that edges are sorted in descending order.

How to Run Python Version

DANI has also a python version. If you want to use the python version, follow the instructions in README.md in python directory.

How to Run Evaluation

Compiling OSLOM2

We use OSLOM2 [2] for the community detection. You have to download OSLOM2 and unpack it in the ./OSLOM2 and compile it to do the evaluation. You can use the script below to do these works.

wget http://www.oslom.org/code/OSLOM2.tar.gz
tar -xvzf OSLOM2.tar.gz
cd OSLOM2
chmod 744 compile_all.sh
./compile_all.sh
cd ..

Note: Compiling OSLOM may result in some errors but it is OK for our usage.

Calculating Metrics

To do the evaluation, run the evaluation code as below.

cd evaluation
python3 main.py --gt <gt> --pr <pr>

Where <gt> is a path to a directory contatining a file named network.txt containing true underlying network and <pr> is a path to a directory contatining a file named network.txt containing inferred network. For example, you can use the following command to run the evaluation on the example dataset.

cd evaluation
python3 main.py --gt ../dataset/groundtruth --pr ../dataset/inferred

Compatibility

This code has been compiled and tested on Ubuntu 22.04 LTS using Python 3.10.

Cite us

@misc{DANI,
    title={{DANI}: Fast Diffusion Aware Network Inference with Preserving Topological Structure Property}, 
    author={Maryam Ramezani and Aryan Ahadinia and Erfan Farhadi and Hamid R. Rabiee},
    year={2023},
    eprint={2310.01696},
    archivePrefix={arXiv},
    primaryClass={cs.SI}
}

References

We have used SNAP library and OSLOM2 to develop this method.

[1] Leskovec, J. and Sosič, R., 2016. Snap: A general-purpose network analysis and graph-mining library. ACM Transactions on Intelligent Systems and Technology (TIST), 8(1), pp.1-20.

[2] Lancichinetti, A., Radicchi, F., Ramasco, J.J. and Fortunato, S., 2011. Finding statistically significant communities in networks. PloS one, 6(4), p.e18961.