POI Recommendation Based On Deep Reinforcement Learning

With the rapid development of wireless communication and mobile devices,such as mobile phones and smart watches,a large volume of data in location-based social networks has been collected in Location Based Social Networks(LBSN)such as Yelp and Foursquare.These tremendous data enable a lot of new research in learning human mobility,e.g.,social relation inference,friendship prediction,etc,among which recommending the point-of-interest(POI)to users they may be interested in but have never visited is one of the most important applications that has received increasing attention from both academics and industry.An extensive number of models have been developed to improve the recommendation performance by exploiting various characteristics and relations among POIs.However,there is very little work to study the mechanism of point of interest recommendation from the perspective of “why users prefer certain points of interest rather than others”.In this work,we initiate the first attempt to use the generative model to simulate the POI recommendation process in the way of generating POI,and learn the distribution of user latent preference by proposing an Adversarial POI Recommendation(APOIR)model,consisting of two major components:(1)the recommender(R)which suggests POIs based on the learned distribution via maximizing the probabilities that these POIs are predicted as unvisited and potentially interested;and(2)the discriminator(D)which distinguishes the recommended POIs from the true check-ins and provides gradients as the guidance to improve R in a rewarding framework.Two components are co-trained by playing a minimax game towards improving itself while pushing the other to the limits.By further integrating geographical and social relations among POIs into the reward function as well as optimizing R in a reinforcement learning manner,we show that our model can recommend the POIs effectively.Finally,we conduct extensive experiments on several real datasets and the results demonstrate that our proposed APOIR model obtains significant performance improvement on all datasets compared to the state of the art baselines.Points of Interest(POI)recommendations are important to both users and businesses.Therefore,research in this field is of great value in both scientific research and engineering.