Research On Semi-supervised Community Detection Based On Constraint Matrix And Linear Representation

In recent years,many scholars have proposed some semi-supervised community detection methods.These methods improve the performance and precision of community detection by combining potentially useful prior information(especially prior information obtained through active learning methods)with the network topology.However,these methods have significant shortcomings in terms of accuracy and efficiency of fusion.Based on these,this article focuses on the shortcomings of the current semi-supervised community detection methods,and improves the semi-supervised community detection methods.The innovation points of the improvements include the following two aspects:(1)To overcome the shortcomings of the existing semi-supervised community detection methods,a semi-supervised community detection algorithm MCSNMF based on a constraint matrix is proposed.The algorithm constructs a constraint matrix to ensure that nodes passing the constraint are eventually divided into the same community and as a result,improving the accuracy of semi-supervised community detection.The algorithm first optimizes the initial adjacency matrix based on the cannot-link constraint information;then uses the must-link constraint information to construct the constraint matrix;after this builds the mapping matrix:maps the constraint set to the community structure;finally uses the non-negative matrix factorization idea to construct the optimization The objective function and the iterative optimization of the mapping matrix to get the final community structure division.The experimental part applies the algorithm and its base algorithm SNMF and the four existing semi-supervised community detection methods Zhang,Zhang?Eh,MMGG?ML,MMGG?ML(C)to artificial benchmark networks and real networks using AC and NMI.The results show that MCSNMF can more accurately mine the community structure.(2)In the process of discovering specific communities,the importance of nodes in the network is different.For some networks,the meaningful nodes are easily known,so the MCSNMF can effectively improve the accuracy of community detection.However,if such nodes are unknown,they need to be mined out in an efficient way.Through active learning,meaningful nodes or node-pair links in the network can be obtained,and then constrain these nodes or links to improve the semi-supervised community detection performance.Therefore,semi-supervised community detection algorithm AL?MCSNMF based on a linear representation mechanism is proposed in this thesis.Compared with the random selection mechanism used in existing methods,AL?MCSNMF uses a node linear representation mechanism,and fuses this mechanism with MCSNMF to make the semi-supervised community detection algorithm more efficient.The algorithm first obtains the ideal topology of the network through a linear representation mechanism;then builds the objective function of the difference between the real topology and the ideal topology,and continuously optimizes the objective function to obtain the final key node;finally,the key nodes are made must-link and cannot-link constraint to get prior information and then MCSNMF.The experimental part applies the algorithm and MCSNMF to artificial reference networks and real networks using NMI and AC as well.The results show that AL?MCSNMF performs better than MCSNMF.