Real Time Recommendation Based On Mobile Context Awareness

In the mobile environment, the user context will be changed with time changes. The recommend system is built based on current context of user. When environment changes, user context will change also, which results in the mismatching between previous recommend result and current context. User models in previous works cannot describe current user context in mobile environment. To deal with this problem, based on the different degrees of context awareness, real time recommendation in mobile context awareness problem is studied in this paper. The main contributions of the paper are as below.(1) We formulate the real time recommendation problem in mobile context.In the mobile context, the context model, matching strategy, and measurement of the recommendation are different from previous works. We first describe the real time recommendation problem in mobile context. After that, we introduce the combination strategy between mobile context and recommendation. Then, we propose real time recommendation problems and sovling frameworks on different awareness levels.(2) We propose the single user oriented real time interest matching strategy in the mobile context.On the single user awareness level, based on the single user context, we introduce the points of interest(POIs) recommendation problems with/without obstacles. Firstly, we define the context elements and the data structures, build the single user oriented context model. Secondly, we propose the Top-k matching algorithms according to the two recommendation problems. Experiment results demonstrate the efficiency of proposed algorithms.(3) We introduce the group user oriented real time interest matching strategy based on social behavior analysis in mobile context.On the group user awareness level, we propose the POIs recommendation problem for group user. Firstly, we analyze the social behavior of users, and then introduce the gourp event extraction algorithm, mining the group information in single user check-in records. Secondly, we build the group user oriented context model through location and best fit scale of POIs. Thirdly, we propose a new data structure combining the context elements, and introduce the Top-k matching algorithm for group users. Experiment results demonstrate the efficiency of proposed algorithms.(4) We propose the community user oriented real time context modeling strategy based on the dynamic community detection.On the community user awareness level, we firstly analyze the mobility behavior of users. By defining the change of history stable contact(CHSC), we find that the CHSC follows the power law distribution in the certain range. Secondly, we propose the dynamic community detection algorithm based on cumulative stable contact among users. Thirdly, we introduce the label based community tracking algorithm, identifying and tracking communities in real time. After that, we also introduce the community oriented recommend process in detail. Experiment results demonstrate the efficiency of proposed algorithms.