Research On Location Recommendation Algorithm Based On Link Prediction In LBSN

With the rapid development of mobile Internet and the popularity of mobile terminal equipment,LBS(Location Based Service)and the traditional social network gradually integrated to form the LBSN(Location-based social network).LBSN used the user's geographical location information to integrate the online virtual social network and offline entity world together,which makes it easier for users to share and get the information they are interested in,and makes more and more users using it.The location recommendation in LBSN not only helps users to find the new location that they are interested in,and helps businesses to brand promotion and precision marketing,which brings huge economic benefits.With the great value of the research,location recommendation in LBSN has become a research hotspot in academic and industry.Although the accumulation of large number of user data from online and offline provides a good basis for the study of location recommendation,since the data in LBSN has the characteristics of large scale,multi-dimension and sparse,which makes that there is still much improvement room on real-time and recommended accuracy against the original algorithms.Aimed at above problems,this paper combines the multi-dimensional information such as time and space in LBSN,and use the complex network link prediction technology to make the location recommendation.The primary work and research results of this paper are as follows:(1)From three aspects about social relations,time and space,we deeply analyzed the users' check-in data of LBSN and then dug out general model of the user check-in behavior.Then,based on the features of LBSN data,we proposed a complex graph model,which contains two kinds of nodes of user and location,and contains three kinds of edges of user-user,user-location and location-position.Meanwhile,this paper proposes a measurement method of the weight of three kinds of edges in graph model.(2)The GraphSF algorithm based on graph is proposed,which reduces the number of user nodes in original graph model.On this basis,a random-walk link prediction algorithm WPPR(Weighted PersonalizedPageRank)is proposed,which uses the link prediction technology of complex network to predict the location recommendation.The algorithm takes the edge weight into account and adds the restart mechanism,which makes it has good recommendation accuracy and running efficiency.(3)Based on GraphX which is a parallel graph calculation framework of Spark,this paper has implemented the proposed algorithm in parallel,which effectively improves the scalability and real-time performance of the algorithm.Finally,under the real Spark cluster environment,we make the comparative experiment with other location recommend algorithms.The results show that the proposed algorithm not only performs well on the accuracy and recall rate,but also has higher efficiency and stronger expansibility.(4)Based on the location recommended algorithms which proposed in this paper,we implement a location recommend prototype system using the technology of Google Map API and Web development technology.