Research On Link Prediction Technology In Social Networks

With the rapid development of network technology and the accelerated pace of people's life,online socialization has become more and more popular among netizens because of its convenient and fast advantages.Link prediction technology has become an important part of social network platform to self-optimization.The increasing users' volume and structure of social network has brought unprecedented challenges to link prediction technology.In view of the low accuracy,slow prediction speed and the inability to discover potential hidden links in the currently used link prediction methods.this paper introduces overlapping community discovery technology into link prediction,and proposes the LPBOCD link prediction model based on an improved q LFMflp overlapping community discovery algorithm.The innovation and work of this paper are mainly in the following three aspects:(1)Because of the low prediction accuracy,long running time and neglecting the discovery and utilization of hidden links of the commonly used link prediction methods,this paper proposed LPBOCD link prediction model based on q LFMflp overlapping community discovery algorithm.The model combined overlapping community discovery techniques with link prediction techniques to effectively reduce the range of link predictions and discover link information between overlapping communities to improve speed and accuracy.(2)In this paper,three improvements of the LFM overlapping community discovery algorithm are made and a q LFMflp overlapping community discovery algorithm that can be quickly partitioned and applied to link prediction technology is proposed.Firstly,this paper proposes to add the "access" flag to the nodes to solve the problem that the same node is repeatedly added and eliminated in the community expansion process of the LFM algorithm.Secondly,for the LFM algorithm results only give the community clustering information unreserved node community attribution information,this paper proposes to perform secondary coding for the nodes,retaining the nodes information while retaining the nodes community information so that it can quickly locate the community where the node is located and apply to link prediction techniques based on overlapping community structures.Finally,it is proposed to use the central node to solve the problem that the LFM algorithm repeatedly calculates the fitness of all nodes in the community every time when one node is expanded.By comparing experiments on the LFR benchmark networks,it is verified that the q LFMflp algorithm has an average increase of 7.3% over the NMI index of the LFM algorithm,and the execution time can be reduced to half,showing a huge advantage in the community partition accuracy and execution rate.(3)This paper designs and implements the LPBOCD link prediction model,performs 10-fold cross-experiment verification on Facebook network data,and evaluates model performance using accuracy,recall rate,F1 index and execution time.Experiments show that the LPBOCD link prediction method improves the precesion by 5%,the recall by 11.4% and the F1 indicator by 8.8% on average compared to the best results of the five commonly used link prediction methods of CN,Jaccard,RA,LP,and Katz(?=0.1).At the same time,compared with the 1 minute and 55 seconds spent on the fastest performing CN indicator,the model only shows a huge advantage in 1 second,which verifies the feasibility and superiority of the model.