This repository contains MATLAB code for 7 multi-view spectral clustering algorithms (and a single-view spectral clustering algorithm) used for comparison in our paper "Multi-view Graph Learning by Joint Modeling of Consistency and Inconsistency", which is the follow-up work of our ICDM paper "Consistency Meets Inconsistency: A Unified Graph Learning Framework for Multi-view Clustering".
The code of some algorithms was gathered from the websites of the authors of the original papers and was later fixed and optimized by us.
Please see our paper for the details of these algorithms (the folder names correspond to the abbreviations for the algorithms in our paper, namely, AASC, AWP, CoReg, MCGC, MVGL, RMSC, and WMSC). In each of these folders, there is a main file xxx_main.m
for the algorithm where xxx
is the algorithm name.
The original papers for the 7 multi-view spectral clustering algorithms and the single-view spectral clustering (SC) algorithm are:
- Huang et al., 2012. Affinity Aggregation for Spectral Clustering
- Nie et al., 2018. Multiview Clustering via Adaptively Weighted Procrustes
- Kumar et al., 2011. Co-regularized Multi-view Spectral Clustering
- Zhan et al., 2018. Multiview Consensus Graph Clustering
- Zhan et al., 2017. Graph Learning for Multiview Clustering
- Xia et al., 2014. Robust Multi-view Spectral Clustering via Low-rank and Sparse Decomposition
- Zong et al., 2018. Weighted Multi-view Spectral Clustering Based on Spectral Perturbation
- Ng et al., 2002. On Spectral Clustering: Analysis and an Algorithm
For the code of our Multi-view Graph Learning algorithm, please see this repository.
All datasets used in our paper are available at Baidu Cloud with code pqti
and Google Drive. Each dataset is a mat file containing 2 variables fea
(i.e., a MATLAB cell of features) and gt
(i.e., ground truth label), except the file flower17.mat
which contains a cell of distance matrices and ground truth since features are unavailable.
- The distance matrices in
flower17.mat
should be squared before passing them into the SGF and DGF functions, and the stringoriginal
should be passed into the functions as the metric parameter. - The datasets
Reuters
,Reuters-21578
,BBCSport
, andCiteSeer
are text datasets with word frequence as features and thus should be used with thecosine
metric for computing distance matrices.
- Windows 64bit:
Add some helper files to MATLAB path by
addpath('MinMaxSelection'); addpath('utils')
command in MATLAB command window. - Linux, Windows 32bit and Mac OS:
Add some helper files to MATLAB path by
addpath('MinMaxSelection'); addpath('utils')
command in MATLAB command window. Then recompile the helper functions by runningminmax_install
.
The file test.m
contains examples to use all the algorithms.