Multi-improvement Of Affinity Propagation Clustering Algorithm And Its Applications

Data mining is a key technology in the era of big data,clustering algorithm plays an important role in the field of data mining,so the research and application of clustering algorithm has theoretical and practical significance.In this paper,the main research direction is the in-depth study of the Affinity Propagation(AP)clustering algorithm.Aiming at the difficulty of determining the Preference of AP clustering algorithm,several theoretical improvements and application researches are carried out from the perspectives of algorithm idea,swarm intelligent,density aggregation and domain intersection,respectively.The specific contents include:(1)Integration Neighbor Points-Based Weighted Similarity Affinity Propagation(IW-AP)is proposed.With IW-AP,the nearest neighbor points are integrated based on the center of the classes,so that the information of the neighboring data points is merged and reorganized,and the range of searching useful information is expanded.The Preference in the traditional AP clustering algorithm is replaced by the weighted similarity degree,which makes Preference more representative.The simulation results show that the clustering accuracy of the improved algorithm is significantly improved compared with the original algorithm.(2)Chicken Swarm Optimization-based Affinity Propagation(CSO-AP)is proposed.The sil evaluation metric is introduced as the fitness value of the chicken swarm optimization algorithm.Through the chicken swarm optimization method,the Preference given randomly in the AP algorithm is iterated.After several iterations,the optimal position of the chicken swarm is given to the Preference in the AP algorithm to complete the optimization.The simulation results show that the improved algorithm is better than the original algorithm.By applying CSO-AP to 34 hotel and tourism companies,the good result of financial data clustering shows that the algorithm has certain practical ability.(3)Density Aggregation-based Affinity Propagation Clustering Algorithm(DA-AP)is proposed.The clustering centers obtained by AP algorithm are recoganized as new dataset,and the idea of DBSCAN clustering algorithm is used to aggregate the high-density neighboring points into new clusters,updating the class labels of all sample points,without debugging the Preference.The simulation results show that the improved algorithm has higher clustering accuracy and better convective data processing effect than the original AP clustering algorithm.By applying the DA-AP clustering algorithm to the foreign exchange market,the good clustering effect of 33 popular exchange rates reflects the practicability of the DA-AP clustering algorithm.(4)Attention-based Affinity Propagation(Attn AP)is proposed.By absorbing the principle of Soft Attention model in Attention mechanism,core sample points that need to be focused on are selected,similarity matrix is obtained,and Preference is optimized to make its value more suitable for data set.The simulation results show that the improved algorithm is better than the original AP clustering algorithm.Through the application of Attn AP clustering algorithm to 5G concept stocks,the clustering results of 47 stock data show that Attn AP clustering algorithm has practical application ability.