Abstract

Abstract. Many real-world data sets, like data from social media or bibliographic data, can be represented as heterogeneous networks with several vertex types. Often additional attributes are available for the vertices, such as keywords for a paper. Clustering vertices in such net-works, and analyzing the complex interactions between clusters of differ-ent types, can provide useful insights into the structure of the data. To exploit the full information content of the data, clustering approaches should consider the connections in the network as well as the vertex at-tributes. We propose the density-based clustering model TCSC for the detection of clusters in heterogeneous networks that are densely con-nected in the network as well as in the attribute space. Unlike previous approaches for clustering heterogeneous networks, TCSC enables the de-tection of clusters that show similarity only in a subset of the attributes, which is more effective in the presence of a large number of attributes. 1