A survey of clustering algorithms based on spectra of graphs

Over the past decade,a huge amount of research has covered the clustering algorithms that are based on the spectra of graphs.It is essential to analyze the relationships among those works so as to reveal the research tendencies.In this paper,the typical works on topics ranging from cost functions to spectral relaxation solutions were investigated and compared in an effort to clearly reveal the essence of these algorithms.Furthermore,the focus was concentrated on several crucial technical issues,including the construction of similarity graphs,the estimation of the clusters' number,the complexity and scalability,and semi-supervised spectral learning.Finally,some open issues were highlighted for future studies,e.g.,finding more theoretical interpretations of spectral clustering,constructing better similarity graphs,selecting features via learning,and the instantiations of concrete fields.