Research On Personalized Location Recommendation Algorithm Based On Time And Tags

With the development of mobile devices, wireless networks and Web 2.0, LBSN (Location-Based Social Networks) have gradually risen, and more and more people like to share and obtain all kinds of information in it. In order to effectively help users quickly find places of interest, personalized location recommendation has become one of the hotspots in recent years. Although there has been research working on location recommendation, some problems still remain unsolved. First, the most general approach for location recommendation is collaborative filtering based on user similarity calculation. However, due to the sparsity of user check-in data, it is difficult to find the similar users. Besides, time and tags are ignored in current research. On LBSN, a user’s check-in behavior has strong correlations with time and tags as the documents can enable location recommendation with more information. Therefore, in this paper, we introduce a novel location recommendation framework based on the time properties and location tags. The main work of this paper is as follows:(1)In this paper, we investigated the time properties of user check-in behavior on location-based social networks. The properties include difference and continuity. Then the two properties will be leveraged to generate a location recommendation framework with time properties.(2)We used LDA to mine the underlying topic based on the tags, and generate the topic-word model of locations. Then we utilize the users’check-in data to build the user-topic model. After getting the two models, ratings will be predicted through calculating the similarity between the users and locations in terms of topics. Finally, we proposed a recommendation framework with time properties and location tags.(3)Last but not least, an experimental study is conducted to validate the recommendation method, using data from the widely used location-based social networks, Foursquare. The information of time and tags include in our model can help us make a comprehensive analysis of relationship between users’behavior and time properties and location tags. Result of the experiment shows that the method introduced in this paper has better performance in location recommendation.