Abstract: Recommendation is one kind of important strategy to promote the active degree of different social networks.However,it is a big challenge to improve the recommendation performance on social networks for the large scale of nodes as well as the complex relationship.Random walk is an effective method to solve such kind of problem,but the traditional random walk algorithm fails to consider the influence of the neighboring nodes adequately.A recommendation method based on random walk on the graph integrated with FP-Growth was proposed,which is based on the graph structure of the social networks.It introduces the FP-Growth algorithm to mine the frequent degree between the adjacent nodes,and then constructs transition probability matrix for random walk computing.Recommendations will be made according to the importance rank of friends.This method not only retains the characteristics of random walk method,such as alleviating the data sparsity effectively,but also weighs the difference of the relationship between diffe-rent nodes.The experimental results show that the proposed method is superior to the traditional random walk algorithm in the recommendation performance.

Key words: Social networks,Friends recommendation,Frequent item mining,Random walk