Research On Point Of Interest Recommendation Algorithm Based On Contextual Information In LBSN

With the popularity of the Internet,electronic devices are becoming more and more comprehensive.In real life,people have their own electronic devices and use them to share their lives.It can be seen that the Internet has become a part of people's daily life.Point of Interest(POI)recommendation algorithm in LBSN allows people to find locations that better match their preferences to meet their specific needs.User check-ins will become more and more numerous over time and the number of users and POI will increase.The explosive growth of information will lead to uneven users visit locations.However,since the user's check-in matrix is sparse,it does not record the POIs users dislike or like explicitly.It is challenging to infer user preferences from implicit feedback data.Second,the contextual information involved in recommendation system contains personal information such as the user's history locations,user check-in frequency and social relationship.The effectiveness of the recommendation may be better with more information provided by the user.Once a third party attacks the recommendation system can cause the leakage of users' personal information.This paper focuses on the POI recommendation algorithm and the POI recommendation with privacy protection.This paper proposes an improved context-aware weighted matrix factorization algorithm for POI recommendation(ICWMF)to improve recommendation performance and have a more reasonable recommendation strategy.It takes advantage of time factor,geographical information,and social relationship to get user's preference for locations.(1)The Ebbinghaus forgetting curve is employed to model the influence of time attenuation to simulate changes in user preferences due to time.(2)In order to assign dynamic weights to unvisited POIs and infer user preference,we build the implicit feedback term by modeling the social relationship and the geographical information from user perspective.(3)From location perspective,the Gaussian model is employed to construct proximity location relationship matrix to represent the probability of locations being discovered by users.Then,we take it as the regularization term to avoid overfitting.(4)The implicit feedback term is used to mine the preferences in the implicit data to assign dynamic weights to the unobserved values.After that,we combine the regularization term to improve the objective function of weighted matrix factorization to recommendation POI.The results of simulation experiments on two datasets indicate that ICWMF outperformed other methods.This paper proposes a POI recommendation method integrating privacy-preserving and social geographical information based collaborative filtering algorithm(PPSGCF)in order to avoid exposing user information when the recommender system is attacked.(1)Integrating social geographical information and collaborative filtering algorithm for POI recommendation(SGCF)is introduced,which considers three factors based on user preferences,social relationship and geographical information.(2)For social relationships we use a differential privacy method to provide users with the ability to protect information.We incorporate a certain quantity of Laplace noise to the similarity between friends for perturbation.(3)From both regional perspective and location perspective,we design a Laplace noise addition method based on different privacy budget allocation weights to protect location information;(4)We incorporate friend relationship and location information with added noise into SGCF to implement a recommendation algorithm with privacy protection.Some simulation experiments have been conducted on the Brightkite and Gowalla datasets.The experimental results show that the proposed method in this paper can effectively prevent the leakage of users' personal information during the recommendation process compared with other algorithms.