Research On Key Algorithms Of Link Pridiction Based On Complex Network Structure

With the fast development of sundry networks such as Internet and social media,the information of the network has shown an explosive growth trend.Link prediction is an important network analysis technology.It aims to predict the potential link relationship in the network through the known network topology so that can help to mine the network structure and evolution mechanism.Since link prediction algorithm using the similarity of network structure has the characteristics of high accuracy and low computational complexity,it has received widespread attention.However,the existing research considers relatively single factors and does not make full use of the message of network structure,resulting in low prediction accuracy.This paper is dedicated to further mining the network topology and studying link prediction on single-layer and double-layer networks.Finally,we propose two link prediction models with high precision.In the research of link prediction in single-layer network,after weighing the accuracy and computational complexity,the paper proposes a node influence index based on the difference of neighbor contributions.This index takes the neighbor contribution into account,and adds a penalty to highlight the influence of nodes themselves,which better measures the influence of nodes.Then,we incorporate the effect of path propagation using a Markov random walk process and use the proposed node influence index to express the information transmission ability of relay node in the path.Finally,an intralayer link prediction model DME is established.We conduct experiments on eight real networks.The results suggest that the predictive precision of DME is 14.6%higher than the model which only uses degree to measure the influence of nodes.In addition,without increasing computational complexity,it performs better than most models and has higher network generality.In the research of interlayer link prediction in double-layer networks,aiming at the problem that the strength of edges before embedding is not effectively expressed,the paper presents a weighted model of edges by integrating the edge betweenness and degree centrality to quantify the strength of edges in the layer.Even in an unweighted network,the strength of relationship between node pairs may not be the same.The higher the relationship strength between nodes is,the closer their embedding vectors in the latent space are.Therefore,the network topology information can be fully utilized by weighting the edges.Then,the network is represented in a low-dimensional dense vector space by an embedding algorithm combining first-order proximity and second-order proximity.Finally,the inter-layer link prediction model SWLEV is established based on the consistency of embedding vectors.We carry out a large number of experiments on real network.The results show that the predictive performance of SWLEV model is outperforms the current progressive models and precision is increased by at least 6.8%.