Abstract:

The affinity propagation clustering algorithm(AP) is sensitive to the preference value, and it is difficult  to find the optimal preference value. 2SAP, a two-stage semisupervised clustering algorithm based on AP, was proposed to overcome this limitation. Semisupervised clustering based on affinity propagation (SAP) was used to cluster the whole dataset and obtain the exemplar set, and then the SAP was used again to cluster the exemplar set to find the final clusters. Experimental results on real data sets showed that the 2SAP was better than SAP and PSAP in terms of CRI and FCRI, and the lower coefficients of dispersion illustrated that 2SAP was less sensitive to the preference value.

Key words: affinity propagation, preference value, semi-supervised clustering, prior knowledge, pairwise constraints