Link Prediction Based On Improved Local Path Index

In the information age,human activities rely more and more on the cooperation between each other,thus forming a variety of intricate relationships and forming different complex networks.Link prediction is an important research direction in complex networks,and it can be used to solve a class of problems which have something in common in many subjects.In link prediction,the link prediction algorithm based on structural similarity has received widespread attention because the network structure is easier to obtain.Among them,the local path algorithm has low computational complexity and good prediction accuracy.However,it is difficult to determine the value range of the variable parameters it uses,and the optimal value cannot be obtained when a fixed value is taken.At the same time,it only considers the number of common neighbors in the second-order path index,and does not consider other common neighbor information.Therefore,this article improves these shortcomings of the local path algorithm and proposes an improved local path algorithm.The research content of this article is as follows:First,the method normalizes the second-order path index and the third-order path index,and defines the normalized second-order path index and the normalized thirdorder path index.After combining these two indexes,the normalized local path index,NLP,is proposed,and the optimal variable parameters of NLP index on different networks are found.The results of experiments show that the NLP index can improve the accuracy of link prediction,and can accurately and simply find the optimal parameter value within the normalized value range,verifying the effectiveness of the new index and the necessity of normalization.Then,in order to improve the prediction accuracy,the local path index and the common neighbor information indexes are combined.On the basis of the third-order path index,the second-order path index is replaced with eight common neighbor information indexes,and the local path index based on common neighbor information,CLP,is proposed.Experiments on different real networks show that the local path index combined with AA index,AALP index,and the local path index combined with RA index,RALP index,perform better than LP index,and can effectively improve the accuracy of link prediction.Finally,the variable parameters in the AALP index and the RALP index are fixed at 0.001,the empirical parameter in the LP index.This makes them unable to find optimal parameters based on the differences between different networks.Therefore,referring to the process of normalizing the LP index,after normalizing the AA index and the RA index,the normalized local path index based on common neighbor information,AANLP index and RANLP index,are proposed.The optimal variable parameter values are found through experiments.The results show that the AANLP index and RANLP index can further improve the accuracy of link prediction.Among them,the prediction accuracy of the RANLP index is higher.