Research On Personalized Location Recommendation Algorithm Based-on Location Based On Social Network

With the ubiquity of the mobile device and the maturity of GPS positioning technology,location-based social networks(LBSNs)become more and more popular.To solve the problem of information overload on LBSNs,researchers focus on finding the locations in which users are interested and providing users with personalized location recommendation service.Collaborative filtering(CF)technique is widely used in recommender system because of its' simplicity,high efficiency and strong interpretability.The recommendation performance of CF depends on the sparsity of users' rating data.In location recommendation,CF is implemented based on userlocation check-in matrix.However,due to few users' check-in records and no negative sample in check-in record,user-location check-in matrix has high sparsity,which leads to bad performance of CF.Therefore,to alleviate the impact of sparsity on location recommendation,this paper further research the location recommendation method.The research work and results are as follows:(1)This thesis proposes a collaborative filtering algorithm based on users' spatialtemporal similarity to alleviate the impact of data sparsity on User-based CF.Based on the periodic influence of time on the user's check-in behavior,the time attribute is introduced by dividing the user-location matrix over time.At the same time,a timesimilarity calculation method is designed,and the user-location-time matrix is filled according to the time similarity,which alleviates the problem of user-location-time sparseness caused by time segmentation.Based on spatial aggregation of user's checkin behavior,the user's spatial similarity is computed by combining the distance between the location and the center of users' active region which are found by multi-center clustering algorithm and users' preference of active region.Finally,the time-aware users' similarity and the users' spatial similarity are combined to obtain the user's spatial-temporal similarity.(2)This thesis proposes a location recommendation method based on users' preference to alleviate the impact of data sparsity on Matrix Factorization algorithm.Firstly,the user's preference for unvisited locations is calculated by combining the user's location type preferences,geographical location restrictions and location popularity.Then a certain proportion of negative samples are selected according to the user preference to fill the user-location 0/1 matrix.Then the weight matrix is constructed according to user's check-in frequency and the user's preference for unvisited location.Finally,a weighted matrix factorization algorithm is proposed and optimized by using the alternative least square method.(3)The relevant comparison algorithms are designed and tested on the real LBSNs dataset,Foursquare dataset.The experimental results show that the proposed method of this paper has better performance than the comparison algorithm.According to the impact of data sparsity on the collaborative filtering in location recommendation,this paper proposes two location recommendation methods respectively,which can alleviate the impact of data sparsity on location recommendation and have certain theoretical and practical values.