The Research On Recommendation Of POI Based On Temporal Geographical And Social Circles

The rapid development of social networks in the information age has become an indispensable part of people's lives.With the extensive application of global positioning system(GPS)and the demand for sharing points of interest,the Location-based Social Network(LBSN)came into being.LBSN attracts thousands of users to register,so the accumulation of massive social data and space-time information on the LBSN site,which provides a good data support for the points of interest.Through the exploration of LBSN data,we find that the social activities of users are timeliness.Timeliness are reflected in the frequency of sign-in,topic of sign-in and social influence,and the user's activity decision is influenced by geographical factors and social factors.In summary,we have the following four points in the paper to explore:(1)We analyze the user's check-in data and social data,and find that the user has obvious timeliness at the check-in behaviors.So the paper divides the original geography circle and the original social circle into timeliness geographical circle and timeliness social circle.(2)This paper explores a variety of geographical factors and social factors based on timeliness geographical circle and timeliness social circle,including: personal preference,proximity of distance,similarity of subject,similarity of,preference,authority and degree of intimacy.According to the conclusion of the related research and the temporal and social data of LBSN,this paper puts forward a scientific and reasonable calculation method for the above factors.(3)GSTS(Geographical,Social,Temporal)is proposed,which is divided into three steps: matrix dismantling,matrix decomposition and matrix merging.First,the matrix is divided into 24 sub-matrix according time interval.Then,the paper decomposes the user-point-of-interest matrix into the user's hidden feature matrix and the point-of-interest hidden feature matrix by using the matrix decomposition technique.Specifically,the decomposition process is transformed into an optimization process to construct an objective function that includes geographical and social factors The result is a result of geographical and social constraints.Matrix decomposition techniques have been widely used in the recommended field,and the recommended results demonstrate that it enhance the recommended performance.And this paper use the matrix multiplication to get the final point of interest prediction matrix,and finally use a variety of matrix merging method for multi-time interval of point of interest prediction matrix.(4)In the final part of this paper,this paper experiment based on the LBSN real data set Foursquare,and the use of accuracy and recall rate as evaluation criteria of the recommended system performance,the experiment is divided into four parts,namely: different authority calculation methods and different matrix merging method and compare recommended performance under the different strategy;the recommended performance under different objective function parameters and compare recommended performance under different parameter;compare recommended performance with other points of interest recommendation;compare recommended performance with the random point of interest recommendation.The experimental results show that the GSTS model can effectively improve the performance of the points of interest.