Clustering attributed graphs: Models, measures and methods

Abstract

Clustering a graph, i.e., assigning its nodes to groups, is an important operation whose best known application is the discovery of communities in social networks. Graph clustering and community detection have traditionally focused on graphs without attributes, with the notable exception of edge weights. However, these models only provide a partial representation of real social systems, that are thus often described using node attributes, representing features of the actors, and edge attributes, representing different kinds of relationships among them. We refer to these models as attributed graphs. Consequently, existing graph clustering methods have been recently extended to deal with node and edge attributes. This article is a literature survey on this topic, organizing and presenting recent research results in a uniform way, characterizing the main existing clustering methods and highlighting their conceptual differences. We also cover the important topic of clustering evaluation and identify current open problems.

Community Detection in Attributed Social Networks: A Unified Weight-Based Model and Its Regimes

“…GAMer [83,87], DB-CSC [86], SSCG [88], FocusCO [166] and ACM [210]. The main idea behind the subspace-based (also known as projection-based) attributed graph clustering is that not all available semantic information is relevant to obtain good-quality communities [84,85], therefore one has somehow choose the appropriate attribute subspace to avoid the so-called curse of dimensionality (see [24,Section 3.2]) and reveal significant communities that would not be detected if all available attributes were considered. 1 Throughout the text, methods and datasets covered by the survey are written in bold.…”

“…There is a variety of surveys and comparative studies considering community detection in social networks without attributes, in particular, [46,69,178,223]. In opposite, the survey [24] seems to be the only one on community detection in attributed social networks. Obviously, since it was published in 2015, many new methods adapting different techniques have appeared in the area.…”

“…Obviously, since it was published in 2015, many new methods adapting different techniques have appeared in the area. Furthermore, a big amount of the methods that had been available before 2015 are not covered by [24], in particular, some based on objective function modification, non-negative matrix factorisation, probabilistic models, ensembles, etc. In a sense, the technique-based classification of attributed graph clustering methods in [24] is also sometimes confusing.…”

“…A reasonable generalisation of the methods of both types are the ones that take into account both network structure and actors attributes. This is a relatively novel direction [24] in social network analysis which is quite promising as simultaneous usage of structure and attributes may clarify and enrich the knowledge about the social actors, to give meaning to the detected communities and describe the powers that form them. During the last decade many methods based on different ideas and techniques have appeared in this direction.…”

“…During the last decade many methods based on different ideas and techniques have appeared in this direction. Although there exist some partial overviews of them, especially in Related Works sections of published papers and in the survey [24] published in 2015, a recent summary of the subject is a necessity as the growing number of the methods may cause uncertainty in practice.…”