HyperGraph Link Prediction

This repo contains codes for link prediction task for hypergraphs. Some codes reproduced from this repo.

Getting Started

Prerequisites

Python version

Clone Repo

    $ git clone https://github.com/alibtasdemir/Hypergraph-Tailed.git
    $ cd Hypergraph-Tailed

Datasets

Used dataset is under this page:

• coauth-DBLP: co-authorship on DBLP papers.
• coauth-MAG-Geology: co-authorship on Geology papers.
• coauth-MAG-History: co-authorship on History papers.
• tags-math-sx: sets of tags applied to questions on math.stackexchange.com.
• tags-ask-ubuntu: sets of tags applied to questions on askubuntu.com.
• threads-math-sx: sets of users asking and answering questions on threads on math.stackexchange.com.
• threads-ask-ubuntu: sets of users asking and answering questions on threads on askubuntu.com.
• NDC-substances: sets of substances making up drugs.
• NDC-classes: sets of classifications applied to drugs.
• DAWN: sets of drugs used by patients recorded in emergency room visits.
• congress-bills: sets of congresspersons cosponsoring bills.
• email-Eu: sets of email addresses on emails.
• email-Enron: sets of email addresses on emails.
• contact-high-school: groups of people in contact at a high school.
• contact-primary-school: groups of people in contact at a primary school.

Download and unzip datafiles in [RAW_DATA_FOLDER]

[RAW_DATA_FOLDER]
 ├── coauth-DBLP
 ├── coauth-MAG-Geology
 ├── coauth-MAG-History
 ├── ...
 └── contact-primary-school

1. Preprocess data files

To preprocess data files use:

python run.py --task preproc --logdir [LOG_DIR] --rawdir [RAW_DATA_FOLDER] --outdir [PROC_DATA_FOLDER]

2. Generate HyperGraph Objects and Save with Pickle

python run.py --task create --hgdir [HG_FOLDER] --graphdir [PROC_DATA_FOLDER] --graphname [GRAPH NAME] --hedgesize [HYPEREDGE SIZE] --sampling [NEG SAMPLING TYPE] --imb [IMBALANCE]

Options

• --graphname: The name of a graph to create pickle object. Pass ""(empty string) to use all graphs inside PROC_DATA_FOLDER. Default: email-Enron
• --hedgesize: The target hyperedge size for link prediction. Choices are [0, 4, 5]. Use 0 to use both 4, 5 hyperedge sizes. Default: 0
• --sampling: Negative sampling technique. Choices are [clique, star, tailed]. Use "" (empty string) to use all techniques. Default: ""
• --imb: Class imbalance ratio (NEG/POS). Choices are [0, 1, 2, 5, 10]. Use 0 to use all. Default: 0

3. Train/Test

To train/test the networks:

python run.py --task predict --hgdir [HG_FOLDER] --resultsdir [RESULTS_FOLDER] --graphname [GRAPH NAME] --hedgesize [HYPEREDGE SIZE] --sampling [NEG SAMPLING TYPE] --imb [IMBALANCE] --feature [FEATURE]

Options

• Check previous section for --graphname --hedgesize --sampling --imb
• --feature: Name of the feature that will be used for representation of nodes/hyperedges. Choices are [gm, hm, am, cn, jc, aa]. Use "" (empty string) to use all. Default: ""
  • gm : Geometric Mean - Wikipedia
  • hm : Harmonic Mean - Wikipedia
  • am : Arithmetic Mean - Wikipedia
  • cn : Common Neighbors - NetworkX
  • jc : Jaccard Coefficient - Wikipedia
  • aa : Adamic-Adar - Wikipedia

Contact

Open an issue for any inquiries. You may also have contact with alibaran@tasdemir.us