Connection Structure Of Graph Clustering Algorithm

Clustering as an unsupervised method is a key attention domain for researchers in the current background and a challenging task.The connection of nearest neighbor points is the key research direction of clustering problem.Data clustering is of great significance to machine learning,information retrieval and data mining,so it has become one of the research hotspots.Clustering problem is the process of making the similarity of the same kind of objects as large as possible,and the different classes of objects as small as possible.In the past decades,graph-based approaches have been increasingly applied to clustering problems.In this method,the similarity matrix of point-to-point is obtained from the original data,and then the similarity between points is obtained.Finally,the clustering results are obtained.The early graph-based clustering algorithm did not take into account the problems in the process of clustering,which seriously affected the clustering performance.Another problem is that the early methods did not utilize the connection structure information of graph,which resulted in a lot of room for improvement in clustering performance and graph quality.Therefore,in this paper,Connection structure of graph clustering algorithm is proposed by combining the connection structure information of graph with clustering problems.Algorithm in this paper utilizes the interconnectivity of points and its nearest neighbors to obtain the subordination of points and centers that determines the clustering results.A point can connect the all relevant nearest neighbors and builds a network,This network is called a fully connected set in the paper,then we replace this point with the center of the network.The networks of adjacent points overlap most with.Hence,the center of networks can decide the clusters.While we optimize the graph model,we also improve the performance of clustering.Experiments on real world datasets and synthetic datasets demonstrate the superiority of our novel approach compared with state-of-the-art clustering methods.