Affinity Propagation Clustering Algorithm Theory Of Improving And Its Application

Clustering is machine learning and pattern recognition field is more active research direction. Affinity Propagation clustering algorithm(AP) is an American scholar Frey et al., 2007, in the journal Science, a new clustering algorithm is put forward. Compared with other clustering algorithm, the affinity propagation clustering algorithm has fast speed, high efficiency, etc., in dealing with large-scale data. But there are still some shortcomings:(1) Can't distinguish the importance between sample feature attribute.(2) Clustering number cannot predict the clustering center before clustering.(3) Cannot be achieved using small samples of tagged, guide the clustering process.In view of this, this article on the study of affinity propagation theory and application research. Empowerment by building characteristics, the introduction of a semi-supervised learning theory, the new intelligent optimization algorithm(such as fruit fly optimization algorithm), and other methods, on the basis of this, the integration of affinity propagation clustering algorithm, puts forward some improved affinity propagation clustering algorithm. In this paper, the concrete research content is as follows:(1) For a detailed analysis and summary to the Clustering algorithm, fruit fly optimization algorithm, a semi-supervised clustering theory.(2) Depth study of the affinity propagation clustering algorithm and clustering process and the application, and summarizes the affinity propagation clustering algorithms of domestic and foreign research present situation.(3) In view of the affinity propagation is unpredictable of the best clustering number, to incorporate fruit fly optimization algorithm and affinity propagation clustering, the affinity propagation based on fruit fly optimization algorithm(FOA-AP) is put forward. Put Silhouette effectiveness index as the fitness function of fruit fly optimization algorithm, to obtain the optimal parameters, optimal clustering performance.(4) Introduce a semi-supervised thought, combined with a small amount of known token information to guide the clustering process, the affinity propagation clustering based on the adaptive feature weighted semi-supervised learning(AFW-SAP) is put forward. Through the experimental simulation results show that the clustering algorithm can significantly improve the clustering accuracy.(5) Build the attribute weights, characteristics of empowerment approach. On the basis of this, combined with affinity propagation clustering algorithm, the variation affinity propagation(VAP) is put forward. The method was applied to 31 provincial government websites in China for clustering and analysis, satisfactory results are obtained.