Discriminating graphs through spectral projections

摘要: This paper proposes a novel non-parametric technique for clustering networks based on their structure. Many topological measures have been introduced in the literature to characterize properties of networks. These provide meaningful information about structural network, but many share similar values given measure [1]. Furthermore, strong correlation between these occur real-world graphs [2], so that using them distinguish arbitrary is difficult practice [3]. Although very complicated way represent and graph, graph spectrum [4] believed be signature [5]. A weighted form distribution spectrum, called spectral (WSD), proposed here as feature vector. vector may related actual structure addition used metric graphs; thus ideal purposes. To graphs, we propose rely two ways project eigenvalues into low-dimensional space. The lower dimensional projection, turns out nicely different classes e.g. from network topology generators [6-8], Internet application [9], dK-random [10]. can advantageously separate would otherwise require complex sets distinguished [9].