Research On Community Discovery Algorithm Optimization Based On Label Propagation

With the development of AI,Blockchain and cloud computing technology,the Internet and the real society are more and more overlapping,and the network life has become a new life form.Community detection can explore the community structure of community network and reveal physical function,potential laws and mathematical significance within the network,moreover provide solutions to their real problems.Therefore,it has been widely concerned by scholars.Community detection has strong theoretical significance and application value in network group identification and community precision marketing.Traditional label propagation algorithm has the problem of high randomness of node update order and node label selection strategy in community discovery.Therefore,based on a large number of researches on node influence and community discovery related literature,this paper takes node influence as the entry point and reduces the randomness of community discovery algorithm by introducing the concept of node influence,thus improving the stability of community discovery.In the field of nonoverlapping community discovery,an improved node-influence algorithm WMC is proposed by using the weighted idea and taking the aggregation coefficient into consideration.Then,the WMC algorithm is integrated into the LPA algorithm,and the LPA-WMC algorithm is proposed.The algorithm uses WMC algorithm to calculate the influence of nodes in the network and arrange them in descending order,so as to determine the initialization order and iteration order of nodes,and proposes the concept of tag influence,which takes into account the influence of tags when there are multiple identical maximum number tags.In the overlapping community area,EKsd node influence algorithm is integrated into COPRA algorithm,and EKCOPRA algorithm is proposed to calculate node influence and label influence according to EKsd value of nodes,so as to reduce the randomness of the algorithm itself.In order to prove the optimization effect of the above improved algorithm,this paper conducted experiments on the artificial network data set generated by LFR and three real community network data sets.The experimental results show that the improved node impact algorithm WMC is highly consistent with the classical algorithm and can effectively identify the core nodes.Compared with the original algorithm,the improved LPA-WMC algorithm and EKCOPRA algorithm get higher modularity and NMI value.It shows that LPA-WMC algorithm and EKCOPRA algorithm can make community partition closer to the standard partition results.The stability and accuracy of the original algorithm are improved.There are two innovations in this paper.Firstly,the aggregation coefficient is integrated into WM node influence algorithm,and WMC algorithm is proposed for community detection.And WMC algorithm is integrated into LPA algorithm,LPAWMC algorithm is proposed,and a new tag selection strategy is proposed to solve the instability of tag initialization phase and tag update phase.Secondly,the EKsd value of the node is introduced into COPRA algorithm,and EKCOPRA algorithm is proposed to solve the problem that the label-membership pairs owned by nodes are less than the threshold value,and the maximum membership degree has multiple random selection problems.On the one hand,because label influence and node influence still have a high probability in large-scale networks,there is also randomness.On the other hand,the community detection in this paper are mainly concentrated in undirected and weightless networks,and the importance of the edges between nodes is treated equally.Therefore,the next research work focuses on community discovery in the right network,and uses the edge weight as one of the measures of node influence to further solve the randomness problem of the algorithm and improve the accuracy and applicability of the algorithm.