Community Detection Algorithm Based On Non-negative Matrix Factorization

Community discovery plays an important role in many research efforts,such as analyzing clustering,system recommendation,and understanding the behavior of complex systems.Scientific research on community discovery has great theoretical significance and practical application value,and has attracted the attention of many researchers from different professional fields.In the research of community discovery algorithms,the popular research directions include dynamic label propagation,modular optimization,clustering and other methods.Community discovery algorithm is a graph based clustering algorithm in essence,many scholars use clustering algorithm to solve community discovery problems.In 2014,Density Peaks Clustering(DPC)algorithm was proposed.The algorithm is a densitybased clustering method,which can identify various shapes of class clusters and easily determine the parameters.But this algorithm needs to set the cut-off parameter manually,and the clustering result is greatly affected by the cut-off parameter.At the same time,the algorithm has high time complexity.Aiming at the disadvantages of cutoff parameters and high time complexity of Density peak Clustering algorithm,a new Density Peaks Clustering algorithm based on node Degree(D-DPC)is proposed.Firstly,when calculating the local density and minimum distance of each node,the node degree is used as the local density of the node,and the number of communities and the list of community core nodes are screened according to the descending order of their product.Then the nodes are allocated and overlapping nodes are distinguished according to membership matrix.Experimental results show that the proposed D-DPC method can accurately divide the number of communities on most networks,and has better performance than other algorithms on some networks.Complex networks are usually represented by adjacency matrix.For complex networks with a large number of nodes,the order of magnitude of adjacency matrix is also very large,and data processing is time-consuming.The non-negative Matrix Factorization(NMF)algorithm provides a new method to deal with large-scale data sets.However,the disadvantage of NMF algorithm is that it requires the number of communities as the input parameter of the algorithm,while the traditional methods to obtain the number of communities are time-consuming and greatly affected by other factors.In order to solve the problem that the number of communities cannot be determined in advance in non-negative matrix decomposition algorithm,a non-negative matrix decomposition algorithm DDPC-SNMF based on improved density peak clustering was proposed.In order to avoid unnecessary calculation work and reduce algorithm time,the graph diffusion method is firstly used to carry out local sampling of the network,and completely irrelevant nodes are screened out.Then,the improved density peak clustering algorithm D-DPC proposed above is used to find the central node list in the data set and obtain the number of communities as the input parameter of the non-negative matrix decomposition algorithm.Experimental results show that the proposed method has good performance in revealing the fuzzy community structure,can accurately divide the number of communities,and the accuracy of community identification is higher than other algorithms.