Research On Community Discovery Algorithm Of Random Walk Based On Structural Activity

In the information age,a lot of information is stored on the Internet.Modeling complex networks can effectively solve various node classification,node clustering,link prediction,influence analysis,visual analysis and other problems.The research of community detection algorithm plays a vitally important role in observing the structure of complex network community,understanding the evolution process of the community and analyzing the prediction of network edge link.With the deepening of research on community discovery algorithms by research scholars,community discovery algorithms based on graph embedding technology,community discovery algorithms based on density clustering,community discovery algorithms based on label propagation,community discovery algorithms based on random walk,community discovery algorithms based on statistical recommendations,community discovery algorithms based on overlapping communities,and community discovery algorithms based on dynamic communities are constantly emerging.In the traditional community discovery algorithm,the spectral bisection algorithm can only divide the network into two communities at a time,and the division result is not good.The GN algorithm proposed by Girvan and Newman does not given the termination condition,and the computation time complexity is high.Because the seed node selection of label propagation algorithms is random and the direction of label propagation is random,the final community division results of such algorithms are unstable.KL algorithm,ASOCCA algorithm,etc.need to adjust algorithm parameters for different data sets,and these parameters have a great influence on the division result,resulting in low scalability of the algorithm.Most traditional community discovery algorithms cannot find overlapping nodes and solve the problem of community division of overlapping networks.Based on the shortcomings of traditional community discovery algorithms,this paper proposes a random walk community discovery algorithm(SARW)based on structural activity and an overlapping community discovery algorithm(SARWCN)based on random walk and central node selection.Three ideas for improving the random walk community discovery algorithm are proposed.Firstly,for the original random walk community discovery algorithm,the problem of ignoring the overall structural characteristics of the community is solved by adding community activity factors,the transfer matrix of random walk is rewritten to improve the modularity of the random walk community division results.Secondly,the merge strategy in the random walk algorithm is improved.Incorporating small communities in the community division results of some datasets can improve the accuracy of community discovery results.Finally,for the problem of community division in overlapping networks,an algorithm for overlapping community division based on random walk and central node selection is proposed.Use the feature vectors of the nodes to find the central node with the largest probability of attraction in the community,thereby discovering the centrifugal node between the two communities.According to the similarity between the centrifugal node and the central node of each community,find the overlapping node and the community to which it belongs,and then divide the overlapping communities.This paper conducts community discovery algorithm experiments on four real complex networks and several artificially synthesized networks constructed with LFR network models.The comparison results of the modularity and NMI of SARW algorithm,SARWCN algorithm and other classic community discovery algorithms on non-overlapping networks and overlapping networks are analyzed respectively.We also use the NetworkX library,iGraph library and Matplotlib library to draw the community division results,and visually display the division results of complex network communities.