In order to provide personal friends recommendation service,social networking sites often recommended friends through registered users’ property or common user neighbor friends. Because of the lack of deep mining to user relationships, the recommendation accuracy was not high. In this paper, formal concept analysis was leveraged to acquire knowledge in data. Two concept lattices were built from the user feature attributes and social networking diagram. The random walk method SRWR was proposed and then the FCASRWR method was put forward with the guidance of concept lattice .The FCASRWR method measured the similarity between users, and recommends friends according to the similarity algorithm to users. The Experiments used Facebook's real datasets , and the experimental result showed that the proposed method had a better performance and proved the accuracy of the method.
