Research On The Point-of-interest Recommendation Strategy Based On Location-based Social Network

The continuous development of Location-based Social Network(LBSN)brings rich resources for users,but it is difficult for users to find information they are interested in.The recommendation system is designed to help users find the information they are interested in and is an important tool for solving information overload problems.POIs(Points-of-Interest)recommendation is a branch of location-based social network.On one hand,POIs recommendation can help users to find some interesting places and rich their experience;On the other hand,it can attract more potential users and raise the popularity of merchants,which brings great benefits to them.This paper first summarizes research status of the point of interest recommendation,and deeply analyzes the advantages and disadvantages of existing interest recommendation algorithm.What's more,it illustrates several typical points of interest recommendation algorithm,and describes the point of interest recommendation influenced by temporal and space features.Then,a hybrid Gaussian mixture cluster recommendation algorithm based on the LFM(Latent Factor Model)is proposed.A detailed analysis of the space features of the user's check-ins found that users always check-in near some "center",on this basis,it respectively introduces the recommendation algorithm based on LFM and recommendation algorithm based on Gaussian Mixture Model.And discusses the flaws of the previous studies,such as the "center" outlier in the process of clustering and need to manually set the number of clusters.We use a greedy EM(Expectation Maximization Algorithm)Algorithm to overcome these disadvantages,meanwhile experiments on both public and real datasets Gowalla and BrightKite prove the validity and efficiency of our algorithm.Finally,a POIs recommendation algorithm combining temporal and spatial features was proposed.It discusses the influence of the point of interest recommendation on temporal features and analyzes the difference and continuity characteristics of user behavior over time.On this basis,the features,according to the data on users sign in each time period,for each period on the sign-in data respectively by using the matrix factorization algorithm is used to estimate sign in popularity.For continuity features,it adopts a continuous period based on cosine similarity smooth processing,this method overcomes the study before in the treatment of the continuity features caused by the data about data sparseness and finally got the point of interest recommendation algorithm based on time factor.Through the integration of spatio-temporal features,a recommendation algorithm for integrating spatio-temporal features is proposed,which improves the recommended accuracy,and proves the feasibility and effectiveness of the proposed algorithm through experiments.