Research On Discovery Of Overlapping Communities In Complex Networks

Many systems in nature can be expressed in the form of complex networks,that is,each entity in nature is abstracted as a node in the network,and the relationship between entities is abstracted as an edge in the network.Complex network is a research hotspot based on interdisciplinarity.A large number of studies have shown that the complex network has a structured characteristic,which is characterized by dense connections between nodes in the community,and relatively sparse connections between nodes in the community.Research on the community structure is helpful to understand the network topology and functional structure,discover the rules contained therein,and provide guidance for the utilization and transformation of complex networks.It is the foundation and key to the analysis of complex networks and has extremely important significance.The community structure is often "overlapping" in the real world,that is,communities are intersected,and many nodes may belong to multiple communities at the same time.The complex network overlapping community structure is more in line with the true appearance of the entity relationship in the real world.For example,most commonly,the network of people in society belongs to multiple social groups,such as family groups,classmate groups,colleague groups,and friend groups.It is of great theoretical and practical significance to study the overlapping community structure of complex networks.The research content of this paper focuses on the discovery of overlapping communities,mainly including the following two aspects:1.Propose an overlapping community discovery algorithm based on the largest spanning tree.This algorithm proposes an edge weight model based on the LFM algorithm.This model considers that the edges in the real complex network may have close connection or credibility.Weights;combined with the maximum spanning tree algorithm to rank the nodes in the network to obtain candidate seed sets.And optimize the initial community division results.This algorithm avoids the random selection of seed nodes by the LFM algorithm,not only can it obtain stable community discovery results,but its performance is better than other representative overlapping community discovery algorithms.2.Propose an overlapping community discovery algorithm based on adversarial generation network,which can complete overlapping community discovery while learning graph representation.For the relationship between the edge and the community,we design and generate an adversarial network model.Through the minimum and maximum competition between the edge generator and the edge discriminator,the alternating iteration training of the two improves their respective capabilities,and finally the generator outputs the node vector representation of the entire network,and The node vector value represents the strength of the node belonging to the community,and the community division result of the entire network is obtained.At the same time,compared with CPM and AGM algorithms,it has high performance.