Research On Structural Hole Mining Algorithm Based On Network Representation Learning

The structural hole theory was proposed to explain how to benefit from the competition of social networks and their cross-relationships.However,with the development of information technology,this theory has been gradually applied to different research fields,such as social network analysis,functional brain network construction,effective isolation of infectious disease prevention and control,etc.At present,the research on structural holes mainly focuses on theoretical analysis and practical application.Compared with this,there are relatively few researches on SH spanners detection methods.The existing classical methods mainly detect SH spanners based on the knowledge of graph theory and combined with the properties of network topology structure.These algorithms are difficult to adapt to large-scale network due to their complex computation and poor scalability.In this work,we further study on the detection method of SH spanners,and the main research results are as follows:1.As the connector between different groups,SH spanners has a special position in the network topology.In order to better study the topological characteristics of SH spanners and improve the identification accuracy of SH spanners.We integrate network representation learning into the SH spanners detection method.By calculating the firstorder harmony and second-order similarity of nodes,on the one hand,the structural information of the network can be better captured;on the other hand,the learning network representation can be better adapted to the structural hole detection.Therefore,we designed a deep SH spanners embedding model that can jointly handle the task of representing learning and structural hole detection.2.Considering that the data in the real network not only has topological structure information,but also has abundant and available feature information,such as the attribute characteristics of user nodes in the social network.We combine the structure and characteristic information of data to deeply discuss the problem of identifying SH spanners,and design a two-channel deep network model.It can complete SH spanners detection while the data feature and structure are combined.In addition,we designed the deliever oprator to guide the unification of the two channels representing learning and the updating of the entire model.Through experiments on multiple real data sets,we proved that the proposed two new identification SH spanners methods can more accurately.