The Application Research Of PageRank Algorithm In The Complex Network Community Detection

In recent years,the development of basic disciplines has spawned many new disciplines,including Machine learning,Deep learning,and Complex network science.Complex network science is composed of Social science,Natural science,Engineering and Computer science.It mainly explores external characteristics of network,describing internal mechanism characteristics,time evolution model and synchronous or asynchronous behavior characteristics of network.Moreover,scholars who examine complex networks have found that most networks have clustering phenomena called the community structure.The community structure in the network may help us understand the information that is hidden in the internal structure,especially for the problem of community division.Due to the complexity of the network,there has not been unified definition of the community structure ever since.Over time,scientists consider that nodes are closely related within the community.Algorithms for community detection published by scholars creatively are more flourishing,which immensely promote the development of Complex network science.However,most algorithms can only work well for a certain part of the network with specific characteristics.In this thesis,we examine a community detection algorithm based on PageRank called PRMCD algorithm.Firstly,this algorithm needs to select some seed nodes within the community.Moreover,we use PageRank random walk to find the nodes closely related to seed nodes to join corresponding seed sets.Finally,the best community structure is preserved through the community segmentation index until there is no community structure in the network.The innovation of PRMCD algorithm mainly has three points: 1)The selection of intermediate vector: the intermediate vector is obtained twice according to the random walk.2)The selection of seed nodes: before the random walk,we need to know the initial distribution of the nodes,it means we need to select the initial seed nodes.The initial seed nodes selected in this thesis is the triangle with sum of the largest degree corresponding to the node with the largest degree in the network.3)The selection of community segmentation index: the community segmentation index in this thesis is determined by the weighted sum of modularity and graph segmentation.In conclusion,we use the modularity index,the NMI index and the ARI index to evaluate the advantages and disadvantages of the algorithm.Meanwhile,we select test networks,including LFR benchmark artificial synthetic network and real world network.We compare our algorithms with previous others,such as LEV algorithm,LPA algorithm and BGLL algorithm.According to the modular index,the NMI index and the ARI index,our algorithm can detect distinct community structure.In addition,the PRMCD algorithm proposed in this thesis is a kind of artificial thinking algorithm,performing well,even in this network that has fewer suspension points.Therefore,many scholars are still exploring new community detection algorithms such as PageRank random block model,hoping to apply new methods to more networks.