Research On Density Peaks Clustering And Application In Community Detection

Most systems in our society can be mapped into complex networks,and analysis for network structure is of great significance to network management.Community structure is an important topological structure characteristic of networks.Community detection plays a major part in the field of complex networks.Community detection is basically a graph-based clustering algorithm.Many scholars have used various clustering algorithms to solve the community detection issues.Density Peaks Clustering(DPC)is a kind of density-based algorithm with the advantages of simplicity and efficiency.However,when solving non-convex data or manifold data,DPC cannot identify the cluster centers accurately due to the density measurement and allocation strategy,causing subsequent allocation naturally becomes invalid.Therefore,in response to these shortcomings,this paper modifies DPC to enhance the performance of the algorithm on the manifold datasets.At the same time,we apply it to the community detection process.The main research contributions of this paper are as follows:(1)Aiming at the problem that DPC is difficult to find the cluster centers when processing manifold data,this paper proposes an improved density peak clustering algorithm based on Jaccard coefficient(DPC-JSLP).By introducing Jaccard coefficient as the new measurement of similarity between samples,the density calculation is updated,so that the algorithm can accurately identify the cluster centers of manifold data.At the same time,a non-central point allocation strategy based on label propagation is introduced,enhancing the robustness of the algorithm.The experimental results on 11 synthetic datasets and 9real-world datasets show that the proposed DPC-JSLP algorithm can not only find the center of manifold clusters effectively,but also explore the hidden patterns and associations between the data.(2)In order to modify the stability of community detection,an improved density peaks Clustering(IDPC)is used for community detection.By introducing the weight of each edge connected to the two points,the adjacency matrix in community detection is converted into the distance matrix required in density peaks clustering,and then the density value of each node in the network is updated based on the Jaccard similarity,and the decision graph is constructed.The selected cluster centers are used as the community center,and each point is allocated to the community where the closest point with higher density is located,and the community division process is completed.(3)In order to further improve the accuracy of community detection,the state transition simulated annealing(STASA)algorithm is used to optimize the clustering results of IDPC.Specifically,three state transition operators are designed: vertex replacement operator,community fusion operator and cross mutation operator.The vertex replacement operator randomly replaces a certain node label with its neighbor's community label value to enhance the diversity of the initial optimization solution,and the community fusion operator replaces all node labels in a certain community with neighbor community label to speed up the optimization process.The cross mutation operator performs cross operation on the high-quality solution set generated by above two operators,and plays a role of local search.Experimental results on the GN benchmark and real-world datasets prove the effectiveness of the proposed algorithm.