In this project, collaborative filtering scheme widely used in recommendation systems is applied to our place recommender system. However, simple collaborative filtering algorithm may not be suitable for proper place recommendation because people have often non-overlapped geographic positions. As a result, the quality of user similarity measuremeThe idea is to summarize the users' check-in spots sharing high-level information (e.g. categories). For example, both "Starbucks" and "Coffee World" are in relation to "Cafe" category. We adopt a three-level hierarchical structure of predefmed categories for venues available in Foursquare database [7]. The bottom level categories assign specific information for venues whilst the upper level categories describe more general information. Figure 3 gives a part of the hieratical structure of categories. In this work, we use only ten top level categories, shown in Table 2, to describe a user interests.nt can be affected [3] [6]. Table 1 shows an example of foursquare check-in spots information. To solve the above problem, we transform users' check-in spots in low-level location space into semantic location space carrying meaningful information of users' interests before calculating the user similarity.