Research On Multiple Points Of Interest Recommendation Models Based On LBSN

As the scale of Location-Based Social Network(LBSN)is expanding,people are willing to check in on it while sharing their location information and social relationships.The check-in behavior on LBSN generates a large amount of location data and social data,which are of great value for users to provide personalized services.In order to make it easier for users to get location information that suits their preferences,Point of Interest(POI)recommendation came into being.The POI recommendation has problems such as sparse data and insufficient personalization,so it has the possibility of improvement in the performance and diversity of recommendation.This paper aims at the above problems to carry out research,the main research contents and contributions are as follows:(1)Aiming at the multivariate factors contained in the user's check-in information,a POI recommendation model based on the fusion of geographical,social and similarity factors is proposed.The model takes into account the distribution characteristics of user check-in records and fully considers the user's current geographical location,it extracts implicit features of users and locations from user check-in records.And then,the specific semantics of social factors are given in combination with the recommendation goals.Based on the historical behavior of users,the cosine similarity is used as the measurement standard to calculate the similarity between users.Finally,the three factors are fused by multiplication to generate an excellent performance POI recommendation model.(2)Aiming at the problem that the influence of time factor is not fully considered in previous POI recommendation model,an improved POI recommendation model with embedded time factor is proposed.This paper analyzes the distribution law of user check-in time period on LBSN and introduces smoothing factor to solve the sparsity problem of user check-in time data,so as to accurately capture the time factor in user check-in information.Based on the consideration of geographical,social and similarity factors,the time factors are integrated into the POI recommendation model to further improve model performance.(3)Taking advantage of the characteristics that users have continuous check-in locations with a high degree of relevance,a new successive POI recommendation model is proposed based on traditional POI recommendation,which integrates user personalized preferences and geographical factors.The user check-in behavior is defined as a fourth-order tensor,and the user's personalized preference is modeled by tensor decomposition to solve the sparsity problem of user continuous check-in data.In addition,the Bayesian Personalized Ranking algorithm is introduced to solve and optimize the parameters of the recommendation model.This paper designs experiments based on the real data set,Gowalla,and compares the three proposed models with baseline and popular methods.The results fully verify the advantages of proposed models.