Research On Point-of-Interest Recommendation Algorithms For Fusion Of Multi-Source Information In LBSN

With the rapid development of mobile devices embedded with wireless communications and location sensors,Location Based Social Network(LBSN)is becoming more and more popular,and Point-Of-Interests recommendations have emerged to help users find their favorite point-of-interests in their active areas.However,with the continuous expansion of information volume and even generates information overload,users have higher and higher requirements for various recommendation performances of the system.How to provide users with more efficient recommendations for points of interest has become a hot spot of current research.The following are the main contents and innovation point of this thesis:1.Point-Of-Interest recommendation is one of the key research of recommendation systems.Traditional algorithms only use user sign-in information for recommendation,and only consider sign-in and no check-in for check-in information,while ignoring the number of user sign-in,user sign-in time factors and Trust relationship between users.In order to improve the accuracy of point-of-interest recommendation,a point-of-interest recommendation algorithm(UGT)combining neighbor selection strategy and trust relationship is proposed.For the check-in information processing,the number of checkin times is used instead of the traditional 0/1 check-in,and time weight is added to the check-in information;for user information,a neighbor selection strategy is proposed to capture user preferences,and only direct or indirect friends of the target user are selected As a neighbor set;for the trust relationship between users,first analyze the attributes of users,then give the calculation method of social status,reconstruct the calculation method of trust degree;finally,a linear combination of multiple factors to obtain the recommended results.By verifying on two real data sets,the results show that the interest point recommendation algorithm has a significant improvement in accuracy and recall rate.2.With the exponential growth of the number of users in the web,most recommender systems are not highly scalable.In order to improve the scalability problem in the recommendation system,a collaborative filtering recommendation algorithm(UBKDT)based on k-d tree is proposed.The algorithm uses a hierarchical spatial partition data structure kd tree to partition the user space of the system according to the user's location,and to verify the correctness of the partition algorithm by measuring the spatial autocorrelation index in each region,and then use the data of the obtained region It is used to predict the number of check-in times of target users on candidate points of interest.The collaborative filtering recommendation algorithm is applied to the check-in data of each sub-region,rather than the entire check-in data,which can greatly reduce the running time of the algorithm.Through experimental comparison,the UBKDT algorithm not only has good recommendation accuracy,but also provides users with faster recommendations and enhances the scalability of the system.