Research And Implementation Of Location Recommendation Algorithm Based Geographical,Social And Categorical Information

Recently,using plenty of multiple data of location-based social networks(LBSN)for recommending users with their preferred locations has become an important feature.Therefore,how to model user's preference toward location with the consideration of characteristics of multiple data is a problem that needs to be solved.In addition,various data in LBSN enable us to model user preference from various aspects and then acquire various recommendation results.So,how to offer fusing recommendation is another problem that needs to be researched.In order to solve those problems,our works and innovations are as follows:1)We calculating similarities between users from various aspect,model personalized similarity for each pair of users based learning to rank.In other words,we consider personalized weights of different similarities for each pair of users.And we propose a new location recommendation approach called USGCFBPR.2)We estimate visiting probabilities for users toward new categories based on A-priori algorithm according to visiting history,predict which categories users may visit in the future based on users' frequencies toward visited categories and users'probabilities toward new categories,and then propose a new location recommendation approach called CPCF.3)We model different recommendation results from different approaches based learning to rank to take full advantage of each approach.And then we combine 1)and 2),propose a new location recommendation approach called GSC F.To evaluate the performance of our approach,we conduct comprehensive experiments using a large-scale public dataset collected from Foursquare.Finally,experimental results show that our approaches offers higher precision and recall than baseline approaches in location recommendation.4)We design and implement a location recommendation system based proposed algorithms that USGCFBPR,CPCF and GSC F,it can offer recommendation lists for user as well as dynamically evaluate the performance of the algorithm.