Research Of Link Prediction Algorithm Based On Similarity In Complex Networks

With the development of computer technology and the continuous evolution of complex networks, the information in complex network has attracted the attention of a growing number of scholars. Usually complex network composed of the object and the relationships between them. The relationships is a collection of links. Link contains large amounts of information in network, so it is necessary to mining links in network. Link mining has many branches, link prediction as an important research content in link mining, mainly used to extract implicit information in network and to add the missing information for the incomplete of data source. Link prediction is to mining upcoming future links or the missing links in network database for some mistakes based on the structure information of the network and node properties.With the emergence of huge amounts of data in complex networks, link prediction algorithms should have a higher request. At present, link prediction algorithms was presented more and more, but its predictive ability needs to be improved. Link predictor based on similarity measure has been the current mainstream research direction for its low complexity and high ability. This paper continued research on the basis of predecessors’ research of link prediction algorithms based on similarity measure. In this study, we use the local information of network to compute the similarity between nodes. Starting from the structural characteristics of complex networks, this paper puts forward two methods FreSim algorithm and MSP algorithm to measure the similarity between nodes.FreSim algorithm is a kind of strong association rules mining link prediction algorithm based on frequent itemsets, the ideas of the algorithm from the Apriori algorithm which is a mining association rules algorithm. The novelty of FreSim method is that it uses the characteristics of strong association rules to filter nodes, and throws away association rules corresponding nodes which does not meet the minimum confidence. With the confidence level to measure the similarity between nodes. Through the experiment of FreSim algorithm we find the prediction performance of FreSim is better than other classical algorithm.At present, SP algorithm based on similarity has the better predict performance while it neglected the endpoint of the path’s contribution to the similarity this phenomenon, this paper put forward another algorithm-MSP. MSP algorithm takes into account the different length paths have different contributions and paths of the same length have different contributions these two factors, and consider that endpoint itself has a certain effect to similarity. Therefore, this paper combined SP and endpoint’s contribution to the similarity and defined the MSP method. Finally, experimental results show that MSP method is better than other algorithms, it has good prediction performance.