Supervised Hyeprgraph Reconstruction (Data and Code)

Anonymous Authors

Data Sources

• Enron: https://www.cs.cornell.edu/~arb/data/email-Enron/
• DBLP: https://www.cs.cornell.edu/~arb/data/coauth-DBLP/ (we use 2016 as training, 2015 as query)
• P.School: https://www.cs.cornell.edu/~arb/data/contact-primary-school/
• H.School: https://www.cs.cornell.edu/~arb/data/contact-high-school/
• Foursquare: https://networks.skewed.de/net/foursquare (the NYC_restaurant_checkin collection)
• Hosts-Virus: https://zenodo.org/record/807517#.YgSOoerMJdg (the data/associations.csv file)
• Directors: https://networks.skewed.de/net/board_directors (the net2m_2002-05-01 collection)
• Crimes: https://networks.skewed.de/net/crime

We have processed all data from the sources and put them in data/. Notice that the P.School dataset is put in (and renamed) school, the H.School dataset is put in (and renamed) school2.

Environment

• python >= 3.5, Anaconda3
• numpy>=1.20
• sklearn>=0.24
• networkx>=2.5.1
• graph-tool>=2.44 (for evaluating the baseline Bayesian-MDL)
• cdlib>=0.2.5 (for evaluating the baselines Demon, CFinder)
• tqdm
• Build: g++ -std=c++11 -pthread cmotif.cpp -o cmotif

Running the Code

python main.py --dataset <dataset> --beta <beta> --features <features>

• <features> can be either count or motif.
• As mentioned in the paper, beta depend on datasets. Combinations:
  • --dataset dblp --beta 1000000
  • --dataset enron --beta 1000
  • --dataset school --beta 350000
  • --dataset school2 --beta 60000
  • --dataset foursquare --beta 20000
  • --dataset hosts --beta 6000
  • --dataset directors --beta 800
  • --dataset crime --beta 1000