摘要:The performance of original affinity propagation (AP) clustering algorithm is greatly influenced by an important parameter: preference (median of similarities between data points), and it may be difficult to identify complex structure data. To address the afore-mentioned issues, this paper proposes two novel methods namely the constraint rules-based affinity propagation (CRAP) and matching micro-clusters hierarchical clustering algorithm (MMHC). The CRAP algorithm can obtain better results by searching the optimal preference value by means of the constraint rules-based search algorithm (CRS). The MMHC algorithm initially takes results of AP as micro-clusters, then they are matched in order to achieve the right partitions of complex structure data. Experimental results demonstrate that the improved clustering algorithm performs better than AP.