Research On Community Detection Algorithm Based On Label Propagation

Community structure detection is a technique for resolving community structures from complex networks using graph topology information,which can be widely applied in social networks,biomedicine,public safety and other fields.Community detection algorithms mainly include modularity optimization,hierarchical clustering,spectral methods and label propagation algorithms,etc.Among them,label propagation algorithms have received extensive attention in the research field because of their nearly linear time complexity and obvious advantages in dealing with large and complex networks,but such algorithms have problems such as high randomness and instability,for which the paper integrates node own information and network topology structure information,two improved community detection algorithms based on label propagation are proposed in the paper.The specific work is as follows:Label propagation algorithm based on node importance.To address the problems such as high randomness of the LPA algorithm and considering the differences among nodes,firstly,the algorithm preprocesses the network,and constructs the descending sequence of node importance.In order to reduce the influence of less important nodes on more important nodes,the non-important nodes with node degree less than 2 are not assigned labels.;then,a new node label update strategy is proposed by measuring the influence of nodes on their neighbors based on node importance and preference;then,node labels are updated based on the new label update strategy,and when nodes have multiple candidate labels,labels are selected based on label weights and preferences to avoid random selection of labels;finally,a post-processing method is proposed to merge communities on demand based on community similarity to further improve the algorithm's accuracy.Label propagation algorithm based on node interaction forces.Based on the above algorithm,the algorithm combines the idea of local optimization of triangular structure to consider the node importance,and to avoid the limitation of the propagation range of nodes with greater importance,the node update sequence is constructed according to the corresponding measurement indexes in the algorithm in ascending order of node importance;then,considering the relationship between nodes and their neighbors and the two-way effect between nodes,the node importance and similarity indexes are used to measure the influence of neighbors on nodes,which is used to calculate the node labeling force;then,the labels are updated based on the neighbor label with the largest force,and when labeling update conflicts occur during label propagation,the highest label weight selection method is used to determine the updated labels,so that the algorithm can detect the high quality community structure.In this paper,a comparative experiment is carried out on artificial network and real network shows that the two algorithms have significant advantages in stability and accuracy.Experimental results show that the proposed algorithm can significantly detect the community structure.