Research On Overlapping Community Detection Algorithm Based On Label Propagation

Step in the era of informationization and intelligence already quietly.The complex network is undoubtedly an important way to study the flow of information.Community detection is an important branch of network science research.It can describe the characteristics of the network,analyze the structure of the network,understand the characteristics of the network,predict the network replacement rules,control the development trend of the network,and guide the decision-making of nodes in the network.The existing literature is mostly found in non-overlapping communities in the network.As the research finds that the nodes are not only the only ones,the research on overlapping community detection algorithms is gradually emerging.The thesis is mainly guided by the idea of label propagation,and through the improvement of the label propagation algorithm to find overlapping communities.Corresponding improvements have been made to the defects of the LPA and SLPA algorithms.Because the general structure of the original network is not clear,the internal chaos of the network is complex,and it is difficult to find the network community.Therefore,this thesis proposes an EPA-based line graph conversion algorithm E-LPA,which can make the network community with the community structure through the proposed line graph.The structure is more obvious,innovatively spreads the label of the side in the form of label propagation,completes the community division of the side,and then corrects the rough community after the label propagation through the objective function optimization,and merges the community by mistake,thereby obtaining the community structure,and The algorithm can detect overlapping nodes,making the LPA algorithm transition from detecting non-overlapping communities to detecting overlapping communities.Aiming at the jitter of SLPA algorithm and the quality of algorithm partitioning,the CSLPA algorithm is proposed.The algorithm constructs the core community by finding the core nodes,and the proposed contribution value index is used to process the unprocessed nodes to form the initial community,and then the SLPA algorithm is used to form the initial community.The correction of the inner node gets the communityresult after the division.Experiments were carried out under artificial datasets and real datasets,with extended modularity and partitioning density as evaluation indicators.It is proved that E-LPA and CSLPA algorithms can increase 10%-60% in EQ compared with eight classical comparison algorithms in terms of community partition quality,and increase by 20%-70% under the PD index,and the results are stable,and the algorithm is robust.