Influence-Based Deep Network for Next POIs Prediction

Abstract

The dramatic proliferation of location-based social networks have resulted in a significant amount of data. This has led to the development of location-based recommendation tools that assist users in discovering attractive Points-of-Interest (POIs). Next POIs recommendation is of great importance for not only individuals but also group of users since group activities have become an integral part of our daily life. However, most existing methods make recommendations through aggregating individual predictive results rather than considering the collective features that govern user preferences made within a group. This insufficiency can directly affect the completeness and semantic accuracy of group features. For this reason, we propose a novel approach which accommodates both individual preferences and group decisions in a joint model. More specifically, based on influencing users in a group, we device a hybrid deep architecture model built with graph convolution networks and attention mechanism to extract connections between group and personal preferences and then capture the impact of each user on the group decision-making, respectively. We conduct extensive experiments to evaluate the performance of our model on two well-known real large-scale datasets, namely, Gowalla and Foursquare. The experimental results show its superiority over the state-of-the-art methods.