CE7490 2019 Fall - Advanced Topics in Distributed System - Project 1: Online Social Network.
This repository is the implementation of "Edge Weight Prediction in Weighted Signed Networks" introduced by Kumar, Srijan, et al., ICDM, 2016.
A list of baseline algorithms can be found in the reference directory, including:
- PageRank
- Bias and Deserve
- Reciprocal
- Signed HITS
- Status Theory
- Triadic Balance
- Triadic Status
Running the experiment requires Python3 with corresponding packages.
The project is supported on Linux and MacOS. It may be possible to install on Windows, though this hasn't been extensively tested.
Anaconda is a library that includes Python and many useful packages, as well as an environment manager called conda that makes package management simple.
Follow the official instrutions of Anaconda. Once it has been successfully installed, run the following command at terminal:
git clone https://github.com/RuihangWang/Graph-Edge-Weight-Prediction.git
cd Graph-Edge-Weight-Prediction
conda create -n myenv python=3.6
To use Python from the environment you just created, activate the environment with:
conda activate myenv
To install required packages and dependencies, run
pip install -r requirements.txt
To see if you've successfully activated environment, try to run 'test.py' with
python test.py
The outputs should be {'1': 0.13333340816497807, '2': 0.08840581470882135, '3': -0.14492748120868965} {'1': -0.3536232588352854, '2': 0.7333331836700439, '3': -1.0}
The standard way to run the benchmark algorithms from command line is:
python ex1.py
python ex2.py
ex1.py is for leaving one edge out prediction of a common set of algorithms.
ex2.py is for leaving N% edge out prediction of a common set of algorithms.
To plot the results, run the following command:
cd plot
python plot.py
Example figures of BTCAlphaNet are shown like follows:
The currently avaliable datasets are stored in '../dataset'. Additional datasets can be downloaded from https://cs.stanford.edu/~srijan/wsn/data/data-wsn.zip