A Point-of-interest Recommendation System Based On Geographical Relationship And Friendship Relationship

The wide spread use of location based social networks has enabled the opportunities for better location-based services through Point-of-Interest recommendation.Personalized location recommendations not only help users explore new places,but also make LBSN more popular and attract more users.In LBSN,geographic influence is a major concern in the Point-of-Interest(POI)recommendation.Because geographic proximity affects the user's check-in behavior.Although the geographical influence on users should be personalized,the current research mainly models the geographical influence of all users' check-in behaviors in a common way,that is the check-in behavior of these users is abstracted into a unified model.However,due to the user's hobbies,work income,length o f leisure time,etc.,the geographical influence of the user's check-in behavior is unique.At the same time,location-based social network contains various forms of friendships,these "friends" can help to provide location recommendations.This paper main ly mines the user's personalized preferences from the geographic relationship based on the personalized user movement track and the friend relationship based on the circle of friends.The main research contents of the thesis are as follows:1)This paper proposes the use of personalized geographic influences for location recommendation,that is,using the kernel density estimation method to model the user's movement trajectory as a personalized distribution,rather than a general distribution for all users.Compared to an isolated check-in location,a user-generated movement track contains a richer set of information,such as the order of access between locations,the travel path,and the stay time of each location.By learning an independent distance distribution from the user's historical movement trajectory according to the kernel density estimation method,the personalized geographical influence of the location trajectory checked by the user on the interest point recommendation is explored.2)This paper proposes to use three forms of friendships(social friends,location friends and neighbors)and user-based collaborative filtering methods to recommend interest points for users.These three friend relationships are pre-defined to represent the user's position in the social network.Aspects of the influencing factors.The friend relationship between users can be used to improve the effectiveness of the recommendation technique.The main idea is to enhance the personalized search and recommendation by using the trust and interest similarity carried between friends.In order to evaluate the effect of the algorithm,this paper uses the Foursquare check-in dataset combined and the current popular algorithm to design a comparative experiment.The experimental results show that the fusion recommendation algorithm based on geographic relationship and friend relationship can effectively improve precision and recall.