Research On Point Of Interest Recommendation On Location-based Social Network

With the rapid development of mobile Internet and the popularity of mobile terminal,Location-Based Social Network(LBSN)are becoming popular,in the same time,the information and the technology which is associated with LBSN,is also been widely used.The popularity of LBSN makes a large amount of check-in datasets.And through these datasets,we can find users' preference,which can be used for Point of Interest(POI)recommendations.Then we talk about the advantages of POI recommendation,it not only can provide help for users' daily life and trip,but also can help users' find new POI;in addition,it provides some suggestions for the related industries.For now,the focus of this subject is how to find out users' preference for better and improve the effect of POI recommendation.Thus,there are two research areas in this thesis,the first is how to improve the combined recommendation algorithm,and the second is how to use the temporal features of users' check-in datasets more wisely.As for combined recommendation,the methods which are based single factor have some limitations in general,however,the excellent combined recommendation methods can enhance the effect of recommendation.First of all,this thesis uses Kernel Density Estimation(KDS)to measure geographical distance among locations,instead of the power-law distribution,then the user's check-in probability can be calculated.Besides,this thesis uses the similarity of users' friends and users' check in locations to measure the impact of users' social relationships.Moreover,user-based collaborative filtering technology is used to calculate user's preferences for POI.Then,we standardize the above factors,and fuse them into a unified framework.As for the temporal features of users' check-in datasets,because of the regularity of user's checkin behaviors,if we can fuse the temporal features and the POI recommendation more wisely,then it can improve the effect of recommendation to some extent.So,this thesis divides users' check-in behavior into different temporal periods,and then,transforms position prediction into probability calculation problem.Meanwhile,because of that user's interest will change over time,so we fuse the influence factor of travel time and the probability calculation,and propose a POI recommendation algorithm combining temporal features.Finally,by comparing with some commonly used algorithms,experiments on a typical dataset show that our algorithms perform better,not the precision only,but also the recall of location recommendation.