To evaluate the system, we conducted an offline experiment on the real check-in sports dataset collected from Foursquare database. We collected 40,000 check-in spots over 10,000 current active users. We then select 20 users and then hide some of their already visited places for examining the recommender method. After that we use the method to recommend top-lO venues to each user by varying the nwnber of K local users. Figure 7 demonstrates the precisions and recalls of top-lO places recommended by the system on different K similar users.