Abstract: With the development of geo-social networks,the practice of utilizing the locations published by GeoSN users to offer them personalized reference services not only benefits users,but also brings the business providers potential profits.As the fundamental enabling technology of the location-based reference services,destination prediction becomes one of the most significant research topics in GeoSNs.Considering the features of GeoSNs,this paper proposed a novel destination prediction method named TraDR specially for GeoSNs based on trajectory decomposition and reconstruction to construct personalized inference model for each target GeoSN user,which not only solves the “trajectory data sparsity problem” faced by common location inference models,but also takes advantages of the rich commonly available public background information.Experiments based on real world dataset were carried out,and results prove the high perfor-mance of the presented method both in prediction accuracy and running efficiency.

Key words: Geo-social network,Destination prediction,Data sparsity problem,Check-in trajectory