Abstract: Recent years have witnessed the tremendous development of social media, which attracts a vast number of Internet users. The tweets these users posted provide an effective way of understanding user behaviors. A large amount of previous work benefits from mining user interest to make friend recommendation. However, the potentially strong but inconspicuous relation between location and interest interaction among social media users is overlooked in these studies. Different from the previous researches, we propose a new concept named neighbor-based friend recommendation (NBFR) to improve the friend recommendation results. By recommending surrounding users who have similar interest to each other, social media users are provided a unique opportunity to interact with surrounding people they may want to know. Based on this concept, we first mine users' interest from short tweets, and then propose to model the user interest with multiple topics under the hypercube structure for friend recommendation. At the same time, we also offer a topic matching shortcut algorithm for more extensive recommendation. The evaluations using the data gathered from the real users demonstrate the advantage of NBFR compared with the traditional recommendation approaches.