Neighborhood-based Hypergraph core decomposition

This repository is the implementation of a PVLDB 2023 paper: "Neighborhood-based Hypergraph core decomposition".

Please cite our extended arXiv version as:

@article{arafat2023neighborhoodbased,
      title={Neighborhood-based Hypergraph Core Decomposition}, 
      author={Naheed Anjum Arafat and Arijit Khan and Arpit Kumar Rai and Bishwamittra Ghosh},
      year={2023},
      eprint={2301.06426},
      archivePrefix={arXiv},
      primaryClass={cs.SI},
      url={https://arxiv.org/abs/2301.06426}
}

Datasets

Our datasets can be downloaded from https://drive.google.com/file/d/12cvz-XtfQUbmj-gqlT9z4DKGMuhqLGjs/view?usp=sharing

• data/dataset/real : real-world datasets (UPDATE 6/10/2023: The earlier version of Aminer dataset had repeated nodes in some hyperedges. A clean version of Aminer has been uploaded)
• data/dataset/synthetic : synthetically generated datasets
• data/dataset/protein: CORUM Protein complex hypergraph
• data/dataset/meetup: Nashville meetup dataset

Format of a hypergraph file (.hyp)

Each line is a hyperedge consisting of comma separated node ids.

Notebooks for plots and case-studies

• notebooks/:
  • PlotGeneration.ipynb: Notebook for plots in the paper
  • python_src/prev_9a.ipynb: Notebook for Diffusion Application plots.
  • densest_subhypergraph.ipynb: Effectiveness analysis & plots for volume densests and degree densest subhypergraphs.
  • casestudyI.ipynb: Notebook for case study I (CORUM Protein complex)
  • meetupCase.ipynb: Notebook for case study II (Meetup hypergraph)

source codes

• cpp_src/:
  • C++ implementation of the E-Peel, Peel, Local-core, Local-core with the optimizations, (k,d)-core, 3 baselines (clique-graph, dist-2 bipartite graph & degree-based algorithms) (algorithms.cpp)
• casestudycpp/:
  • C++ implementation of Densest subhypergraph extraction (densest_subhypergraph.cpp)
• par_src/:
  • OpenMP implementation of Parallel local-core (parallel_localcoreinitOpt2.cpp)
• python_src/: Python code to run Application I (Diffusion).

Output

output/: Output folder that contains algorithm outputs.

How to run:

• Sequential Algorithms: Peel, E-Peel, Local-core, Local-core+OptI, Local-core+OptI+II,Local-core+OptI+II+III and Local-core(opt):
  • cd cpp_src
  • bash run.sh // Deactivate(Activate) logging with log=0(log=1), set #runs of an alg. with var it
• Parallel Local-core algorithm:
  • cd par_src
  • bash hgrun.sh // Deactivate(Activate) load-balancing by setting lb=0(lb=1)
• Application I (Diffusion):
  • cd python_src
  • bash run_diffusion.sh

C++ compiler requirements:

• C++11
• OpenMP 3.5

Python and notebook requirements:

• python 3.8 or above
• dependencies:
  • disjoint-set (pip install disjoint-set)
  • hypernetx (pip install hypernetx)
  • pandas (pip install pandas)
  • numpy (pip install numpy)
  • seaborn (pip install seaborn)
  • matplotlib (pip install matplotlib)