Research On Network Structure Optimization And Community Detection Based On Adaptive Link Selection

Complex networks can not only visually display the interaction of various elements in complex systems,but also realize in-depth analysis of the structural features and laws of complex systems through computations.Therefore,they are widely used in various disciplines and fields.In recent years,more and more attention has been paid to the research of complex networks,especially community detection.Mining the structure of the network community is conducive to observing the global information and network characteristics of the complex network.It is an important way to understand the structure and function of the entire network.Therefore,the accuracy of community detection is very important.However,most of the current community detection results are not ideal.The reasons can be roughly attributed to two aspects: First,the community structure of many complex networks is ambiguous,as well as the difference between the number of edges within a community and the number of edges between communities is very small.It leads to insufficient internal connections within the community and is difficult to discover the ideal community structure;second,the actual network often has multiple connection types,and there are obvious differences and commonalities between them.Existing studies mostly regard such networks as simple and ordinary network to conduct community detection.It ignores the global connection information of the network.In order to solve the above problems,this paper conducts the following research based on the network structure optimization and community detection of adaptive link selection:(1)Due to the widespread presence of noisy links in the real network,the community structure in the network may be very ambiguous.This problem directly leads to the decline in the performance of community detection tasks.Therefore,effective screening of network links has become one of the key tasks of community detection.In response to this problem,this paper proposes an adaptive link selection method that can accurately eliminate noisy links in the network and extract effective links in the network,thereby improving the accuracy and robustness of community detection.This method designs a robust matrix decomposition framework that can simultaneously capture sparse noisy links(cannot-links)in the network and structured low-rank effective links(must-links)with smooth edge similarity.In order to improve the computational efficiency of the algorithm,this paper also designs an effective iterative optimization scheme.According to the comparison experiments results on multiple baseline datasets,it can be concluded that the proposed method can greatly achieve pretty performance of the community detection.(2)There are multiple interaction relationships between network nodes,which makes the community structure of the network significantly different in different connection types.Currently,the use of multi-layer networks to simulate multiple types of connections has become a new trend in the research of complex networks.The existing research on multi-layer network community detection usually assumes that each layer of the network has a potentially consistent topological structure,and its purpose is to learn compatible and complementary information from different layer networks,so it can dig out shared global structural information.However,due to the existence of noisy links,this assumption does not hold true in many real networks.In response to this problem,this paper proposes a multi-layer network structure optimization algorithm based on cooperative low-rank representation and sparse decomposition.The algorithm can find the complementary structure information between layers of the multi-layer network and refine the synergy between different layer networks.And decompose them into robust consistency representations to mine the multi-layer network community structure.In addition,we also designed an effective iterative algorithm to optimize the model.It is shown that the proposed method can greatly improve the accuracy of the community detection of multi-layer complex networks according to the comparison results on multiple real datasets.