Geographic Point Of Interest Recommendation Based On Deep Neural Network

The Internet has flourished for decades,and the Location-based Social Network(LBSN)has gradually penetrated into everyone's life.Point-of-Interest(POI)recommendation has become more and more important.By capturing the user's real preferences to recommend POIs that users may be interested in,not only can the user be facilitated,but also bring huge economic benefits to the business.In LSBN,users check-in their favorite POIs and generate additional information.Therefore,incorporating the relevant information into the POI recommendation is the key to improving the performance.However,previous methods mostly ignore the influence of friends on users when recommending new POIs to users,and friend-based collaborative filtering does not take into account the user's own preferences.Then we propose FSNR,a hybrid recommendation model based on deep neural network for the Top-N POI recommendation.The FSNR model integrates the historical check-in records,social relationships and the geographical location of POI,learns the influence of friends from three aspects and models the user's potential preference for POIs.Specifically the FSNR model first models complex user interactions through deep neural network techniques and latent factor model,learning the potential representation of users and points of interest from implicit feedback.The FSNR model fuses social relationships with geographic information to derive three similarities between the user and their friends: check-in similarity,social similarity,and location similarity,and the user's decision is affected by the similarity between users.Finally,the final score of POI is generated by the user's preference score and friend's suggested scores,and the first N POIs with highest score are selected and recommended to the user.Experiments on three real data sets compare the FSNR model with other methods by precision and recall and analyze the influence of model parameters.The experimental results demonstrate that the FSNR model has better recommendation performance.