Analysis Of Successive Point-of-Interest Based On Recurrent Neural Network

With the rapid development of information technology and mobile Internet,the number of mobile devices is increasing.Location-based Social Network(LBSN)becomes more and more popular,for example Foursquare and Gowalla.Point-of-Interest(POI)recommendation plays a critical role in LBSN.Since it attracts wide attention on industry and academia,which become an important research focus.Traditional POI recommendation is based on user interest,and ignores the temporal relationship between user's behaviors.Users usually care more about the location which they are interested at the next moment.Predicting the user's interest in the next moment will help users make decisions,on the other hand,it can help merchants to estimate the number of users at different times and helps businesses to make reasonable arrangements.This recommendation method is called successive POI recommendation.Therefore,we can believe that successive POI recommendation has higher research value than traditional POI recommendation.However,successive POI recommendation still faces with some challenge.First,user-POI matrix is very sparse.That is to say,every user may visit a few POIs,which this area contains a large amount of POIs.Secondly,better modeling sequential behavior of users is hard.In order to solve the above challenge,this paper designs two novel models to improve recommendations and user experience by mining temporal and geographic information.Specifically,the main work of this article is as follows.1.Successive POI recommendation based on fine-grained interest of the user.Successive POI recommendation need to focus on the order of POI which users visit and model the sequential behavior of user.To solve above challenges,this paper proposes a Neural Network model based on fine-grained interest.Our proposed model considers the user interest,POI preference and textual information synchronously.Specifically,in order to obtain fine-grained interest of user,this paper splits the user interest into two factors,namely long-term interest and short-term interest.2.N-successive POI recommendation based on attention mechanism.Successive POI recommendation doesn't take the specific time into consideration while recommending the POIs.If a given future time interval is too long,it may be insufficient for successive POI recommendation to suggest only one POI because visiting that single POI may only consume a short time.Motivated as such,this paper extends successive POI recommendation,called N-successive POI recommendation.And then,present a neural network models for N-successive POI recommendation.The model incorporates personal interests,location preferences,and time factors,and makes recommendations built on user needs in real time.Then it uses an attention mechanism to focus on the latest POIs to capture short-term interest.Above two models are validated by Foursquare and Gowalla dataset.The Foursquare data contains 342,850 check-ins made in Singapore between Aug.2010 and Jul.2011.The Gowalla data contains 736,148 check-ins in California and Nevada between Feb.2009 and Oct.2010.The first model recommends POI set for each user at the next time without considering specific time.And the model is evaluated by precision and recall.The results demonstrate that the proposed model has better performance than the comparison algorithm.The second model specifies N time periods and recommends a POI for each time period.When N equals to 1,N-successive POI recommendation and successive POI recommendation are equivalent.The results of recommendation is evaluated by precision and recall which consider POI order.The evaluation found that the model has a good recommendation compared with other comparison algorithm.