Personalized Point-of-interest Recommendation Based On LBSN

In recent years,with the popularity of mobile terminal devices with built-in GPS(such as smartphones and i Pads),location-based social networks(LBSN)have achieved unprecedented development,such as Foursquare and Gowalla.For users,these websites not only provide them with a platform to share location information,but also can find more point-of-interests(POIs)that they may be interested in through the historical check-in information of other users.This can encourage and help them explore more interesting unknown places.For service providers on the platform,they can mine users' preferences from the users' historical check-in information to make more accurate service recommendations.Therefore,the research of POI recommendation based on LBSNs is of great significance to both individual users and service providers.Although the current POI recommendation based on LBSNs has been studied by many scholars,most of them are score predictions on users' visit preferences.However,users really care about the ranking of the first few POI recommended by the system.Therefore,the POI recommendation is actually a ranking problem rather than a score prediction problem.Based on the theory of Learning To Rank,this thesis conducts related research on the POI recommendation based on LBSNs.The main research contents and results are as follows:1.In order to solve the problem of inadequate utilization of missing check-in data by traditional recommendation methods,this thesis proposes a POI recommendation algorithm based on geographical influence and extended pairwise ranking by studying the spatial distribution of user history check-ins.The algorithm first performs cluster analysis on user check-in information,obtains all the users' active areas,and combines geographic factors and check-in frequencies to build user preference measurement model.Second,it proposes a learning framework of extended pairwise ranking based on active areas,and combines preference measures model for POI recommendation.Finally,experiments on two data sets show that the algorithm proposed in this thesis has a good recommendation effect.2.In order to solve the problem that the low-scoring dimension will be compensated by the high-scoring dimension when the traditional matrix factorization model reconstructs the user check-in score.This thesis proposes a non-compensated ranking POI recommendation algorithm based on context weighting.The algorithm first mining the impact of social relationships and geographic factors on users' check-ins through collaborative filtering and adaptive bandwidth kernel density estimation.Secondly,it proposes a non-compensated matrix factorization model that reconstructs user preferences based on non-compensated rules,and combines the model with weighted bayesian personalized ranking method to recommendation.Finally,experiments on two data sets show that the algorithm proposed in this thesis is better than the comparative methods.