Research On Personalized Recommendation System Based On LBS

With the fast development of Location-based service(LBS), some problems and challenges are springing up. How to clear the barrier and make use of them to find more development chances and opportunities? Researchers and practitioners have developed a lot of frameworks and applications. Personalized recommendation may be the most effective method. The reasons are the three following statements:First, more and more e-commerce activities are extending to mobile-internet, it needs personalized recommendation. Second, LBS is developing faster and faster, it has to face information overload sooner or later and personalized recommendation can handle this well. Third, the base of LBS is a mobile device owned by individual, so it can represent and must deal with individual character better than e-commerce, at which personalized recommendation is good.As a result, LBS needs personalized recommendation and can provide more personalized data and information for personalized recommendation. But the method for incorporating personalized recommendation into LBS is not mature yet. Because the incorporation raise three challenges:First, LBS needs more accurate service because of its location-awareness, but the traditional personalized recommendation deal with e-commerce unrelated to location. Second, the request for LBS always burst out and not continuous, people have to figure out how to process interest drift under LBS environment well. Third, the application area of LBS are much wider than e-commerce, so the cold boot problem, a long existing in personalized recommendation field, becomes more serious.This thesis raised a LBS-based personalized recommendation framework against the above problems and challenges. The framework is consisting of several parts:a long-term interest model for stable and common user interests and a short-term interest model for variant and special user interests. Context information modeling and filtering is applied in both interest model and the difference is that what used in short-term interest model is pre-filtering and long-term interest model is post-filtering. In order to overcome the cold boot problem, this thesis try to apply multi-criterion method in short-term interest recommendation and accumulate the information and data for making use of content-based recommendation in long-term interest recommendatio.In the final part of this thesis is a case of the framework, the data from www.dianping.com is employed as the empirical test and the result shows that the framework can somewhat solve the three problems above mentioned.