Joint cluster analysis of attribute and relationship data withouta-priori specification of the number of clusters

摘要: In many applications, attribute and relationship data areavailable, carrying complementary information about real world entities. such cases, a joint analysis of both types can yield more accurate results than classical clustering algorithms that either use only or (graph) data. The Connected k-Center (CkC) has been proposed as the first cluster model to discover k clusters which are cohesive on However, it is well-known prior knowledge number often unavailable in applications community dentification hotspot analysis. this paper, we introduce formalize problem discovering an a-priori unspecified context data, called X Clusters (CXC) problem. True assumed be compact distinctive from their neighboring terms internally connected Different attribute-based methods, neighborhood not defined but To efficiently solve CXC problem, present JointClust, algorithm adopts dynamic two-phase approach. phase, find so atoms. We provide probability for thisphase, gives us probabilistic guarantee, each true represented by at least one initial second these atoms merged bottom-up manner resulting dendrogram. final determined our objective function. Our experimental evaluation several datasets demonstrates JointClust indeed discovers meaningful clusterings without requiring user specify clusters.