Research On The Combination Clustering Method Based On Choice Preference And Implementation

Grouping is a process of dividing the individuals in a collection into several subsets with independent significance based on a certain rules. Each group contains one or more individuals. Grouping the consumers in e-commerce context is to divide the service consumers into several groups based on their preference in selecting products or services provided by service providers. The clustering algorithm is one of the effective ways to divide the consumers into groups. At present, a single clustering algorithm is mainly used to resolve the problem of grouping, but each single clustering algorithm has different disadvantage. At the same time, serial computation of the clustering algorithm may take a huge amount of time when processing the large-scale data. Therefore, the following respects were completed in the thesis:(1) Analysis on the related single clustering algorithmsThe idea and the implementation steps of K-means clustering algorithm and affinity propagation clustering algorithm were are introduced in detail, and the advantages and disadvantages of both the algorithms were analyzed while they were applied in dividing the consumers into groups.(2) Design of the combined clustering algorithmsTwo kinds of combined clustering algorithms are presented, including the combined clustering algorithm AAK (Affinity Propagation+Affinity Propagation+ k-means) and the combined clustering AKK algorithm (Affinity Propagation+ k-means+k-means). The idea and implementation steps of the AAK algorithm and the AKK algorithm were introduced in detail. The superiority of the combined clustering algorithms, without the prior knowledge on the number of clusters and with high accuracy of division, was emphasized.(3) Implementation of the MapReduce parallel models of the combined clustering algorithmsThe high order MapReduce sequence linking technology was introduced. And then the parallel models of the two kinds of combination clustering algorithms were built by using the technology, which achieved the high efficiency of the combined clustering algorithms.(4) Experimental verification of the combined clustering algorithmsThe serial model and the parallel model of the AAK algorithm and the AKK algorithm were implemented on platforms Matlab and Hadoop respectively. Through comparing the experimental results with that of other combined clustering methods, the advantages of the AAK algorithm and the AKK algorithm were validated.