As an important clustering algorithm, Affinity Propagation (AP) algorithm can quickly find the reasonable clustering center. But the AP algorithm is difficult to make correct clustering, when the sample in the weaker separability feature space. In this paper, the Domain Knowledge Blended Affinity Propagation (DKB-AP) algorithm is proposed. Combining the domain knowledge function and the similarity measure of the AP algorithm, the algorithm makes iterating to obtain the clustering result. The experimental data are three random sample sets, including two sample sets whose subclass aggregation degree are good, one sample set whose subclass aggregation degree is weak. The clustering results, Fowlkes-Mallows Validity Index and Error Ratefor for the Sets are analysed. The results show that the clustering result in the weaker separability feature space by DKB-AP algorithm is almost consistent with the clustering result in the separability feature space by AP algorithm.
Chen, Wei; Tian, Qichong; Jiang, Xiaorong; Tang, Zhibo; Guo, Caihua; and Xu, Xinzheng
"Domain Knowledge Blended Affinity Propagation,"
Applied Mathematics & Information Sciences: Vol. 07
, Article 39.
Available at: https://dc.naturalspublishing.com/amis/vol07/iss2/39