A social network comprises a set of users, called nodes, connected with each other by different relationships, called links, such as friendship, job, religion, common interests, information sharing, etc. Link prediction, which is to infer new links among nodes in a network should exist or are likely to occur in the near future, is a key research directions in social network analysis. Classical approaches to link prediction has tree problems: 1) no effective way to combine network structure and content feature; 2) not deal with non-transitive similarity; 3) not suitable for large-scaled networks. We want to deal with these problems by proposing a new link prediction method. This proposed method will use the multi-hashing idea to capture the similarities of the nodes’ network structure and content features, but also put some constrain on them. The result of our synthetic data experiment shows the proposed method is promising in capturing community structure, which will be used for link prediction purpose.