| As an important branch of complex network research,link prediction has important application value in different disciplines.Link prediction in complex networks predicts and restores unknown links and future links in networks based on known network information.In the context of big data,complex networks have the features of diverse types,complex structures and large data scales.Therefore,how to fully mine the known structural information of the network for high-precision link prediction has become a new difficulty.At present,the link prediction algorithm based on naive Bayesian theory is mainly faced with the problems of insufficient network local information mining and the difficulty in meeting the independence assumption.In order to solve these two problems,this thesis makes the following contributions in the field of link prediction from the perspective of motif structure information and common neighbor nodes packet:Traditional prediction algorithms based on naive Bayesian theory only focus on the local structure information of common neighbors,ignoring the contribution of motif structure in the network to link formation.Aiming at the problem of insufficient local information mining in naive Bayesian algorithm,a motif-based naive Bayesian link prediction algorithm is proposed.Firstly,the motif density is quantified based on the aggregation of motifs on the path structure.Then,the role contribution function of differentiated path contribution is defined based on the motif density.Combined with naive Bayesian theory,a naive Bayesian link prediction adapting to the network with motif structure is proposed.Finally,the proposed algorithm is compared with three types of algorithms based on local information,semi-local information and global information.The results show that the proposed algorithm has better prediction effect on the network with obvious motif features.In the process of improving the independence assumption,the existing hidden naive Bayesian algorithm and tree augmented naive Bayesian algorithm focus on calculating the similarity contribution of the correlation between common neighbors,ignoring the fact that the closely related common neighbors and relatively independent common neighbors affect the formation of links at the same time.In order to solve the problem that the independence assumption in the naive Bayesian algorithm is difficult to meet,the hybrid naive Bayesian link prediction algorithm based on packaging is proposed.Firstly,this thesis uses the conditional mutual information between common neighbors to calculate their correlation degree,and designs the packaging criterion according to the correlation degree,so that the common neighbors are divided into the correlated common neighbors and the independent common neighbors.Then,the above two naive Bayesian algorithms are improved by applying the packaging criterion,and two hybrid naive Bayesian algorithms that calculate the contribution of the correlated common neighbors and the contribution of the independent common neighbors are obtained.Finally,the experimental verifications are carried out on the real network with high aggregation and rich common neighbors.The results show that the naive Bayesian algorithms based on the common neighbor packaging are not only superior to the original algorithm without packaging,but also have good robustness. |
