Exploiting multi-attention network with contextual influence for point-of-interest recommendation

Abstract

Point-of-Interest (POI) recommendation has become an important service on Location-Based Social Networks (LBSNs). In order to improve the performance of recommendation, besides the check-in data generated in LBSNs, researchers are striving to exploit various auxiliary information such as social relation among users and geographical influence among neighbourhood POIs. However, existing works cannot effectively study the diverse degrees of influence from user’s friends, neither are they able to capture the feature impacts of POIs in the preference modelling process. To overcome these challenges, by making use of a M ulti-A ttention N etwork to learn the C ontextual influence of both users and POIs, this paper presents a model named MANC for POI recommendation. The MANC model consists of two parts: a user-friend module and a POI neighbourhood module. Unlike existing works which treat the influences from different friends of a user equally, the user-friend module in MANC applies an attention-based memory component to generate specific relation vectors which can differentiate the influence from the aspect of interest, and applies a friend-level attention network to adaptively capture the preferences of users. For the POI contextual information, the POI neighbourhood module in MANC applies a feature-level attention network to capture the latent features of neighbourhood POIs, and applies a POI-level attention network to capture the geographical influence among POIs. Extensive experiments are carried out, and it is shown that the MANC model achieves better performance than other state-of-the-art methods.