DPR-Geo: A POI Recommendation Model Using Deep Neural Network and Geographical Influence

In the wake of developments in artificial intelligence, deep learning technology has been used in location-based social networks (LBSNs) to provide web services that meet the needs of users. Point of interest (POI) recommendation, as one of the most important mobile services, aims to recommend new satisfactory POIs to users according to their historical records. However, existing models that uses original high-dimension user vector or location vector cannot capture useful information from historical records effectively. Meanwhile, most of them complete recommendation service only in terms of user’s perspective or location’s perspective. Hence, in this paper, we propose a novel deep learning framework for POI recommendation. Firstly, we use a multi-layer neural network to reduce the dimension of user vector and location vector. Then, we construct a union neural network by concatenating and multiplying vectors to obtain the preferences of users. Finally, considering the unique geographical characteristic of location, we model the distance probability to enhance recommendation. Experimental results on real-world dataset demonstrate our model outperforms some popular recommendation algorithms and achieves our expected goal.