Model selection for semi-supervised clustering

摘要: Although there is a large and growing literature that tackles the semi-supervised clustering problem (i.e., using some labeled objects or cluster-guiding constraints like \must-link" \cannot-link"), evaluation of approaches has rarely been discussed. The application cross-validation techniques, for example, far from straightforward in setting, yet problems associated with have to be addressed. Here we summarize these provide solution. Furthermore, order demonstrate practical applicability methods, method model selection based on this sound procedure. Our allows user select, available information (labels constraints), most appropriate (e.g., number clusters, density-parameters) given problem.