Clustering Method Based On Graph Data Modeling And Reliability Detection

Most traditional clustering algorithms do the analysis and calculation directly in feature space and can not achieve an intuitive and effective graph visualization for the clustering result of high-dimension data.And for the clustering algorithm based on distance similarity,it is easily to understand and effectively to calculate but it can not effectively handle the data sets with non convex distribution in space.In this paper,a clustering method based on the model of graph data is proposed,which reflects the data in feature space into graph space and clusters.then obtains the clustering result of original data.At the same time,a testing method about reliability based on Page Rank Algorithm is proposed in this paper to test the reliability of the clustering result.Firstly,the researching status,significance and thinking of traditional clustering algorithms and graph-clustering algorithms is summarized to prepare for subsequent research.Secondly,some relevant concepts of data mining technique are described briefly and based on them,the classification and relevant concepts of traditional clustering algorithms are introduced further.At the same time,the relevant concepts of graph(network)and complicated network are introduced too.And the classification and relevant surveys about graph-clustering(community mining in network)are illustrated detailedly.Thirdly,according to the traditional problems of the clustering algorithms in feature space,and combining with the experimental analysis of different data sets,a model constructing method about graph data is proposed,and does the clustering analysis based on modular measure to the graph data gathered from model them obtains the clustering result of data in original feature space.Fourthly,the concept of “reliability of adjacent boundary” is proposed.Thinking of the deficiency to put the soft clustering result of fuzzy clustering algorithm as the measurement of reliability and the characteristics of graph data,a testing method about reliability based on Page Rank Algorithm is proposed.Finally,we compare the simulation results of open-source dataset based on the method we proposed with the traditional clustering method.We also developed a comprehensive experimental platform based on MFC,which can load the data and do the relative experimental analysis.