Deep Learning Model For Next POI Recommendation With Uncertain Check-Ins

The research recommended by the next point of interest has great application value for location-based service providers and users.The current research on the next point of interest is based on the user's certain check-ins,ignoring the uncertain check-ins.But at present,mobile data often has the problem of uncertain check-ins,and this article studies this.In order to solve the problem of uncertainty check-in in the recommendation of the next point of interest and poor recommendation performance,this paper analyzes the impact of multiple context information on the recommendation result in the next point of interest recommendation,and combines multiple context information with deep learning technology Two models of recommendation for the next point of interest are proposed: a point of interest recommendation model based on LSTM fusion of multiple features and a point of interest recommendation model based on LSTM interactive MLT.First,for the problem of insufficient information mining in the next point-of-interest recommendation problem and the inability to obtain the potential relationship between consecutive sign-in behaviors,this paper proposes a point-of-interest recommendation model based on LSTM fusion of multiple features.Using fuzzy features to indicate uncertain check-in,the next point-of-interest recommendation model that combines multiple contextual information is combined with LSTM to capture the user's continuous check-in behavior sequence,learn the user's behavior patterns,and then predict the next point of interest the user will visit.Secondly,in order to further improve the prediction ability of the next point of interest and solve the communication problem between multiple tasks,this paper designs an interactive multi-task point of interest recommendation model based on LSTM,which is used to learn the points of interest and the points of interest.interaction.The point-of-interest recommendation algorithm for interactive multi-task learning introduces:(1)a time-conscious activity encoder to reveal the transfer characteristics of categories;(2)a space-aware position preference encoder to capture the transfer characteristics of points of interest;(3)specific The task decoder uses the learned latent transfer features to interactively complete the category and point of interest prediction task.Finally,the two models proposed in this paper are verified on three real-world data sets,and compared and analyzed with 7 comparison algorithms and 8 variant algorithms.