Location-based social networks (LBSNs) provide a platform for users to share their location information with each other. Location recommendation is the task of suggesting unvisited locations to the users. It aims to make satisfying recommendations of locations by utilizing the information such as users' visiting histories, user profiles and location profiles.
This thesis investigates the utilization of check-in data and location category information for location recommendation on LBSNs. A distributed crawler is developed to collect a large amount of check-in data from Gowalla for the research. Then, three ways are used to utilize the check-in data, namely, binary utilization, FIF utilization, and probability utilization. According to different utilizations, different Collaborative Filtering recommenders are introduced to do location recommendation. Experiments are conducted to compare the performances of different recommenders using different check-in utilizations. Location category information is utilized for location recommendation by considering the temporal and spatial patterns. A user's periodic check-in behaviors at different location categories are represented as temporal curves. A temporal influence model is used to predict similar users' check-in behaviors based on temporal curves. A geographical influence model is proposed to filter out locations that are not of interest to the user. By integrating temporal influence and geographical influence a location recommendation algorithm called sPCLR is proposed to recommend locations to the users at a given time of the day. Experimental results show that the sPCLR algorithm performs better than three existing location recommendation algorithms.