Link Prediction Algorithm Based On Community Relevance

With the rapid development of mobile Internet technology,complex network with data mining as the core has become a hotspot of current research.As an important method of complex network research,link prediction based on community structure has attracted extensive attention of scholars at home and abroad.However,the link prediction algorithm based on community structure has some problems such as coarse granularity of link prediction accuracy and ignoring the impact of community role information on community correlation.This paper conducts in-depth research on link prediction between complex networks from two dimensions of community division and improvement of community correlation.Firstly,aiming at the problem that the prediction accuracy of link prediction algorithm of community structure is coarse and the similarity index of node path is easily affected by network structure,a community correlation link prediction algorithm(SMDA)based on multi-granularity partition is proposed.In this algorithm,a multi-granularity community division method was designed to fully consider the results of multiple community divisions under different granularity,so as to avoid the impact of inaccurate community division on the prediction performance.Based on this,relevant indicators of community were set according to the community path similarity information,and community correlation model(CRM)was constructed according to different weights of community structure information divided by different granularity.Community correlation matrix was calculated according to THE CRM model,and link prediction was carried out by combining the matrix with node attribute information.Accurately predict the probability of missing links in the network.Secondly,the existing link prediction algorithms based on community structure ignore the impact of community centrality on the correlation between communities,resulting in the same influence of communities in different roles,a link prediction algorithm based on community centrality and community correlation(SCCA)is proposed.Given community centricity related theory,the algorithm based on correlation theory and community node number relationship is constructed based on the community of centricity correlation model(CCM),according to CCM model to calculate the correlation degree between any two communities,the classical similarity index is applied to the node link projections for the network in the model.Finally,SMDA algorithm and SCCA algorithm are used for link prediction in real data sets with different network characteristics,and compared with classical link prediction algorithm.The results show that the prediction effects of the two algorithms are improved in multiple data sets,SMDA algorithm is more suitable for fuzzy network,SCCA algorithm is more suitable for clear network,and SCCA algorithm is more operable than SMDA algorithm.