Research On Link Prediction Algorithm Based On Matrix Factorization

The network is an effective form of abstractly describing complex systems that exist widely in the real world.The nodes in the network represent various entities in a complex system,and the links in the network represent various connections between entities.Faced with increasingly large network data,the network structure will continue to evolve and become difficult to control.How to accurately predict the missing links of the network and better grasp the evolution trend of the network topology is an urgent problem that needs to be solved in the field of network research at present,that is,link prediction.Link prediction has been widely used in many fields such as communication networks,biological networks and recommendation networks.This thesis mainly studies the link prediction from the two aspects of noise and dynamic evolution of the network.The main research contents are as follows:1.Regarding the noise of the network,most of the current researche on link prediction is carried out under the premise of assuming no noise,ignoring the credibility of the network.In this thesis,a method for prediction of network link with noise based on matrix decomposition is designed.Due to the existence of various noises in different networks,this thesis uses LRR to filter the noise caused by the false information of the network.Secondly,the network is repeatedly perturbed by superposition to overcome the noise caused by the randomness of the network.Finally,by combining LRR and NMF to cope with various noises in the network,effective link prediction is performed in the noisy networks.2.Regarding the dynamic evolution characteristics of the network,the current research mainly focuses on the network topology and ignores the problem of the impact of the dynamic evolution characteristics of the network on link prediction.This thesis designs a dynamic network link prediction model based on matrix factorization.Due to the dynamic change of the network,the dynamic network can be regarded as a collection of multiple static networks through time slicing.The network can be obtained by sequentially predicting the network at each moment in the network collection by using the improved NMF added to the timing characteristics.Change trend,so as to complete the link prediction of the dynamic network.This thesis uses real network data to experimentally verify the proposed model and method.The experimental results show that the model proposed in this thesis effectively utilizes the network topology information and network temporal evolution characteristics,the model can accurately predict the network structure in complex situations,which is helpful for the network data mining and utilization.