Algorithm Research For Link Prediction Based On Weighted Link And Data Field

With the development of information technology, a large scale network data was produced. It makes the analysis on large-scale network becomes possible. In recent years, network mining have a rapid rise and become a very popular research branch. Link prediction is an important branch of network analysis, it is a new challenging research direction. This paper focuses on link prediction problem. On the basis of the existing link prediction algorithms, focuses on the similarity between the node link prediction algorithm.Through long time study,we find out that current approaches of link prediction based on nodes similarity ignore the strength of the link, and the weight setting for the paths based approach is not intuitive. According to these problem mentioned above, a link prediction algorithm based on data field and weighted graph is proposed. This method assign different weight for each link according to the topology of the graph, and take into account interaction between potential links, finally calculate the similarity with data field potential function Experiment on some real-world network show that new method generally improved prediction accuracy and more easy to determine parameter, have higher practical applicability. For dynamic networks typically vary, usually in large-scale and high real-time requirements in applications but existing algorithms are complexity and high costs in updating, it is difficult to achieve real-time requirements, this paper proposed a specific storage structure for network and incremental link prediction methods,so as to achieve the purpose of updating the network quickly.