Continuous POI Recommendation Based On Convolutional And Recurrent Neural Networks

Internet data brings us more timely and broader information,but also faces information overload and quality differences.For example,when we search for a place on the "Mei Tuan",there will be dozens of places,so it is difficult for a user to make a choice,and point of interest(POI)recommendation in a location-based social network(LBSN)is important because it helps the user navigate multiple candidate POIs and provides the best POI based on the user's most recent registration information.However,the location-recommendation is different from the product recommendation.The product recommendation does not need to consider the distance and time too much,as long as the user likes it,but the location-recommendation has strict requirements on time and location,the location is not suitable and the time is unreasonable,and it is difficult for users to reach the location recommended by the system,Location users are difficult to reach,or a user likes fitness.You recommend him to the gym for an unreasonable recommendation in the middle of the night.On the other hand,the location-based social network data has a large difference between the number of POIs and the total number of POIs.The collaborative filtering recommendation algorithm user-POI data is very sparse and the classic collaborative filtering recommendation of method relies on the problem manually extracting data features with very limited validity and extendibility.Based on the above considerations,this paper proposes the recommendation of continuous POI,aiming to integrate the time and geographical factors into the recommendation system,and combine the user's historical check-in data with the current time to make personalized recommendations for users.In this paper,on the other hand to users,each sign in time,the relationship between the POI and sign in and sign in the relationship between the mapped to a tensor,using the depth of learning through an end-to-end process data characteristics,extract the data characteristics of the deeper,the classic collaborative filtering recommendation method depends on prior knowledge to extract data features and user-POI data extremely sparse problem.Considering that the user check-in relationship is long-term dependence,CNN convolutional neural network is good at extracting potential features,and LSTM long-term memory model is good at solving the problem of long-term dependence,so CNN and LSTM are combined to extract features in the experiment.T time of the given user in the experiment,according to t-1,t-2...The check-in data at time t-n can predict the longitude and latitude of the check-in at time t.By querying the POI longitude and latitude table,the first five POIs which are nearest to the user are calculated as recommendations,so as to achieve the purpose of recommending the system according to the user's current time and history check-in record.Finally,experiments on two real data sets,Gowalla and Foursquare,show that the performance of our PONET model is better than previous work.