Research On Adaptive Point-of-interest Recommendation Approachs Based On User Check-in Behaviors

Point-of-interest(POI)recommendation in location-based social networks(LBSNs)can help users effectively explore unvisited locations according to their check-in preferences,and can solve the problem of location information overload by providing personalized POI recommendation service,which has important research significance and application value.However,existing POI recommendation methods have some deficiencies in the aspects of how to deeply study the diversity of users'check-in behavior,how to realize the self-adaptive POI recommendation algorithms combining temporal and spatial features,how to realize the successive adaptive POI recommendation algorithm utilizing recurrent neural network,and how to improve the performance of POI recommendation algorithms,etc.To address these issues,this paper focuses on user check-in behavior and adaptive POI recommendation algorithms,and the main researches are as follows.Firstly,aiming at the shortcoming of less consideration of the user check-in behavior characteristics and lack of user classification in current POI recommendation,a feature extraction method for user check-in behavior and a user clustering algorithm are proposed.The three-dimensional features of user check-in behavior are extracted and analyzed from LBSNs historical check-in datasets using probabilistic statistical analysis methods,i.e.,the check-in number of the user,the number of check-in locations of the user,and the check-in time distribution of the user.On this basis,a user clustering algorithm based on fuzzy C-means is proposed,and then the performance of clustering is analyzed by three indices.After determining the optimal number of clusters,the results of user clustering and the practical implications are given.A thorough and comprehensive study of user check-in behavior and user clustering is realized,which lays a solid foundation for the design of adaptive POI recommendation algorithms in the next step.Secondly,aiming at the problems that user-based collaborative filtering can not accurately acquire the time characteristics of users'check-in preferences and lack of adaptive recommendation strategy,an adaptive POI recommendation method based on temporal features and user-based collaborative filtering is proposed.According to the extraction and analysis of time variability and correlation features,the time slots smoothing technique is fused in the calculation of user similarity,which reflects the users check-in time preferences in collaborative filtering and solves the problem of data sparsity.According to the extraction and analysis of user similarity,a similar user neighbor selection mechanism based on threshold filtering is proposed,which effectively alleviates the computational burden of user similarity and improves the tightness of relationship among users.For users with different check-in characteristics,the corresponding strategies are proposed to recommend POIs,and the adaptive algorithm is realized.Thirdly,aiming at the problems that single recommendation method of spatial probability models and the weak correlation with user'check-in time preferences,an adaptive POI recommendation method is proposed,which combines temporal-spatial features and probability models.Based on the feature extraction and analysis of POI popularity,a method for calculating the popularity of POIs is proposed,which combines the overall effect and time-aware popularity.By modeling and analyzing the distance features of adjacent checked-in POIs,a large number of POIs in LBSNs are filtered and preprocessed to obtain candidate POIs set,which can improve the spatial correlation of POIs and reduce the complexity of recommendation calculation.By choosing two spatial model strategies in the POI recommendation algorithm,namely the time-aware one-dimensional power law function probability model and the time-aware two-dimensional Gauss kernel density estimation probability model,adaptive recommendation for users with different check-in behavior characteristics is realized.Fourthly,aiming at the problem that existing successive POI recommendation methods can not accurately reflect users'short-term interest preferences and can not be applied to users with inactive check-in behavior,an adaptive successive POI recommendation method based on long-term and short-term interest preferences is proposed.In this method,the user's check-in POIs trajectory sequences are divided into short-term check-in trajectory sequences and historical check-in trajectory sequences according to the time window,and the recurrent neural network model and long short-term memory network model are used to model and obtain the user's short-term and long-term interest preferences.For users with inactive check-in behavior,a check-in trajectory sequence filling method is proposed,which fills the recent check-in records of active similar users into the recent check-in trajectory sequence of inactive users.Using recurrent neural network model to model the sequence,the interest preferences obtained are regarded as the short-term interest preferences of inactive users.This method can effectively solve the problem of short continuous check-in trajectory sequence and cold start for inactive users.Finally,in order to verify the effectiveness of the proposed adaptive algorithms,several POI recommendation algorithms are experimented on Foursquare and Gowalla datasets,and the performances are compared and evaluated according to the recommendation precision,recall,F_?index,time average absolute error,running time and training scalability.Extensive experiments show that the proposed methods outperform the baseline methods.