Research On Link Prediction Approach Based On Matrix Factorization

Link prediction is a basic and crucial research in data mining;It is utilized to forecast the unknown links stem from information loss or possible links between two nodes in the future by the known information such as network topology and node properties.Link prediction has not only crucial scientific significance for the analysis of network evolution and data completation through network loss research,but also has important application in the friend recommended,e-commerce,biological pharmacy and so on.Recently,the link prediction has attracted widely attention in many fields such as computer science,statistical physics and biology,a lot of link prediction algorithms that derive from various subject perspective are reported.The matrix decomposition is a predicted method that solve the link prediction by low-rank approximation of the adjacency matrix of original network in the existing prediction techniques.The existing methods on account of matrix decomposition are mostly based on the network model of adjacency matrix,and the modeling of network is still not accurate enough.For this purpose,in-depth study on the link prediction algorithm based on matrix decomposition is conducted in this paper.The main contents include the following two aspects:(1)A matrix decomposition link prediction algorithm was proposed based on the important degree of edge in this paper.In this paper,a link prediction model based on network weight matrix decomposition is established to measure the importance of the connected edges in the network from the perspective of node importance.The algorithm can improve the accuracy of link prediction to certain extents comparing with the existing matrix decomposition prediction algorithms in eight open network data sets.(2)A matrix decomposition link prediction algorithm based on aggregation coefficient is proposed.The algorithm is an important degree prediction matrix decomposition algorithm in edge nodes gathered on the basis of further consideration factor influence on even the edge between node important degree,to establish the link prediction model based on network weights matrix decomposition.Experimental results show that the model can obtain better link prediction results.Two new predictive algorithms of matrix decomposition links are proposed to solvethe problem that the existing matrix decomposition link prediction algorithm is too strong in this paper.The prediction accuracy can be improved to a certain extent,which further enrich the system of link prediction based on matrix decomposition.