Research On Weighted Network Community Division Method Based On Label Propagation

The relationships between many things in real society can be abstracted into complex networks.Community structure is an important characteristic of complex networks,and community partitioning is an effective way to understand network structure and explore the information contained in the network.At present,research on community partitioning algorithms both domestically and internationally is mostly based on undirected and unweighted networks.Compared to unweighted networks,weighted networks can reflect deeper information,and studying the community partitioning problem of weighted networks has broader application value.And directed networks are also an abstraction of the objective world,and the identification of their community structure also has important practical significance.The label propagation algorithm is a simple and efficient community partitioning algorithm with high reference value.Therefore,this thesis researches the weighted network community partitioning method based on label propagation,and the specific work is as follows:A weighted network community partitioning method SLWCD based on similarity and label propagation was designed for undirected weighted networks.This method follows the idea of the LPA algorithm that is using labels to represent the communities that nodes belong to and achieving community partitioning through propagation(iterative update)of labels.However,considering the characteristics of the weighted network,and the randomness in the initial label allocation and label update process of the LPA algorithm may result in different division results(i.e.not stable enough),a new initial label allocation strategy and a new label update strategy have been adopted in the SLWCD method.The initial label allocation strategy introduces a Jaccard based weighted network node similarity that covers both direct and indirect relationships between nodes,while the label update strategy considers edge weights that reflect the strength of relationships between nodes.The results of community partitioning on real Zachary Karate club networks,Lesmis networks,and LFR artificial synthetic networks have verified that the SLWCD method can not only accurately partition the weighted network’s communities,but also has high stability and low time complexity.A directed weighted network community partitioning method ILDWCD based on improved label propagation was designed for directed weighted networks.This method still divides communities through label propagation,but the initial labels are not randomly assigned,but rather introduce similarity suitable for directed weighted networks,group nodes based on node similarity,and then assign initial labels according to the group;During the label propagation process,an improved k-shell algorithm is used to calculate the node importance and determine the label update sequence based on the node importance.The results of community partitioning on artificial synthetic networks validate the effectiveness of the ILDWCD method.In order to verify the application value of the ILDWCD method,a simple micro-blog user recommendation display system was developed using the Spring Boot,Spring MVC,and Mybatis frameworks to demonstrate the process and effectiveness of micro-blog user recommendation based on the ILDWCD method.