Research On Link Prediction Algorithms Based On Matrix Decomposition

Link prediction in complex networks mainly uses the existing link information between nodes and the attribute characteristics of nodes to estimate the probability of links between any two nodes in the network at the next moment.It is usually used to mine the potential or wrong link information in known networks.In recent years,link prediction algorithm has been widely used in many fields,for example,recommendation tasks,public opinion detection system,and the problem of link prediction has been the focus of scholars' attention.In this paper,two new link prediction models are proposed based on matrix decomposition to overcome the shortcomings of existing prediction algorithms that can not fully reflect the structural characteristics of the known network.By extracting the internal structural correlation and external text correlation between the nodes in the known network,better prediction results are obtained.Firstly,a new link prediction algorithm based on matrix decomposition(LPMF)is proposed.LPMF takes the structural information of the known network as input to train the vector representation of nodes in the network in the process of network representation learning.Secondly,the LPMF algorithm is further improved,and a link prediction algorithm based on text enhancement(TELP)is proposed.TELP algorithm takes the text characteristics of external nodes into account in the process of network representation learning,which can more effectively and accurately mine and extract the internal structure correlation and external text correlation between nodes in a known network.Finally,through simulation experiments on three real citation network datasets,the final prediction results are compared with more than 15 existing classical link prediction algorithms,and ROC regional curve(AUC index)is used as the evaluation criterion of the accuracy of prediction algorithm to test the feasibility and accuracy of the LPMF algorithm and TELP algorithm proposed in this paper.The experimental data show that the performance of the LPMF and TELP algorithms in these three datasets is better than that of the other 14 prediction algorithms except the matrix forest index(MFI)algorithm.In summary,the two new link prediction algorithms proposed in this paper can combine the internal structure information of the known network with the text characteristics of the external nodes in the process of representation learning to achieve joint learning of the target network,and show good results in link prediction tasks.