Research On Fuzzy Clustering Algorithm Based On Modularity

The modularity can be used to detect the strength of community network structure.The larger the value is,the closer the inner relationship inside the same community is,and the more distant the relationship between different communities is,which coincides with the goal of clustering.Therefore,more and more clustering algorithms are designed based on modularity in recent years,and the better clustering results have been achieved.However,most of these algorithms belong to the category of hard clustering,limited by the membership degree that can only be 0 or 1,the accuracy of clustering can be further improved.In addition,after the process of clustering analysis,cluster validity test is also a key link.It is often used to measure the performance and accuracy of clustering algorithm by the value of validity index,and the proposed validity index or judgment result is not accurate enough,or has poor robustness.In response to the above problems,this paper mainly studies the fuzzy clustering algorithms based on modularity,the research work is as follows:(1)In order to improve the accuracy of clustering algorithms,A fuzzy co-clustering algorithm MMFCC(Modularity Maximization based Fuzzy Co-Clustering)based on maximization of modularity is proposed.The fuzzy idea is applied to the hard co-clustering algorithm CoClus(Co-Clustering).The fuzzy membership matrices of object and attribute are calculated alternately until the modularity increases to a stable state.That is to say,the difference between the values of modularities produced by the two iterations is less than the threshold value,which leads to the optimal clustering results,and the result clusters are displayed as diagonal matrix blocks.In this paper,we use sparse real datasets to carry out experiments.The experimental results show that compared with other diagonal co-clustering algorithms and fuzzy co-clustering algorithms not considering modularity,MMFCC algorithm not only improves the clustering accuracy,but also represents the meaning of co-clusters intuitively and clearly.(2)In order to verify the accuracy of the clustering results of fuzzy clustering algorithm and integrate the principle of bipartite modularity,a fuzzy clustering validity index,CSBM(Compactness Separateness Bipartite Modularity)is proposed,whichcombines intra-cluster compactness with inter-cluster separateness.The application of bipartite modularity can avoid falling into the local optimal problem.The introduction of intra-class compactness and inter-class separateness will help to enhance the robustness of the index for evaluating the clustering results.In addition,the index can improve the problem that the comparative indices present a linear trend with increasing of cluster number.In this paper,the fuzzy clustering algorithm is carried on six real datasets,and the results show that,The CSBM index is superior to the other six fuzzy clustering validity indices in terms of accuracy of the results and robustness.