GLSUR: POI Recommendations Based on Similar Users and Current Geographic Location

With the rise of location-based social networking, predicting users' future POI(point of interest) has also become a key issue. Now there are some methods that can predict the next point of interest based on the user's own history hidden state. Despite this, these approaches take only into account the user's own behavior trajectory, ignoring the impact of similar users. We design a new recommendation model GLSUR to improve the accuracy of the predictions and provide users with a better experience of the actual situation POI recommendations. The three factors of hidden state, similar users and user embedding are considered simultaneously in GLSUR. We invoke RNN networks as the underlying computational model to calculate the user's own hidden state, the similarity formula determines the similar users, and pass the three tensors into the fully connected layer. Finally, a semi-positive vector formula is used to calculate the closest point of interest and recommend it to the user. We conducted extensive experiments using two real data sets in the real world. The results show that GLSUR provides better accuracy compared to other baseline methods.