stefanresearch / signed-oracle

signed-oracle

This is the implementation for the paper

Sublinear-Time Clustering Oracle for Signed Graphs by Stefan Neumann and Pan Peng, ICML'22.

Oracle Implementation

The implementation of our signed oracle can be found in folder "implementation". To use our oracle, you first need to compile it, e.g., using:

g++-11 -O3 -Wall -fopenmp -std=c++11 oracle.cpp -o oracle To compile it, you need a compiler that is compliant with c++11 and you need OpenMP if you want to use parallelization.

To run the oracle, use the following call:

./oracle inputFilePath numSteps numWalks

The parameters of the oracle are as follows:

• inputFilePath: The path to the file that stores the graph. We assume that the graph is stored using a sparse format, in which each edge has the format "u#v#sign" where u and v are the vertex IDs and sign is the edge sign. We assume that sign is either 1 or -1.
• numSteps: The number of random walk steps to be used by the algorithm.
• numWalks: The number of random walks to be used by the algorithm.

Optional Oracle Parameters

Additionally, the oracle accepts the following optional parameters:

• --seedCluster ... endSeedCluster: Allows to specify a new ground-truth seed cluster. For example, "--seedCluster 1 30 48 70 --endSeedCluster" creates a ground-truth cluster of seed nodes containing vertices 1, 30, 48, 70. It is important that the cluster is ended with "endSeedCluster".
• --randomSeeds s: Tells the oracle to randomly sample s seed nodes and to preprocess them as described in the paper.
• --clusterAll: Returns a clustering of all vertices in the graph. Calls the WhichCluster-procedure for each vertex in the graph.
• --clusterVertices ... endClusterVertices: Returns a clustering of the given vertices. For example, "--clusterVertices 5 10 39 4588 endClusterVertices" returns a clustering of the vertices 5, 10, 39, 4588. It is important that to include "endClusterVertices".
• --numEstNorm r: Specifies how many times the algorithms runs the estimateNorm-procedure. Default: r=1. If r>1, the algorithm runs estimateNorm() r times and then return the median.
• --unsigned: Runs the unsigned version of the oracle (see paper).
• --biclustering: Runs the biclustering version of the oracle (see paper).
• --sameCluster u v: Returns whether vertices u and v are in the same cluster.

Running the Competing Methods

To run the competing method polar by Xiao et al. (WebConf'20), download their code and put it into /include/signed-local-community-master. The code can be downloaded from the following link:

https://github.com/xiaohan2012/signed-local-community

To run the competing method FOCG by Chu et al. (KDD'16), download their code and put it into /include/KOCG.SIGKDD2016-master. The code can be downloaded from the following link:

https://github.com/lingyangchu/KOCG.SIGKDD2016

Running the Experiments

To run the experiments, execute implementation/main.py.