GraphSAGE-based POI recommendation via continuous-time modeling

With the proliferation of Location-based Social Networks (LBSNs), user check-in data at Points-of-Interest (POIs) has surged, reshaping user-environment interaction. However, POI recommendation remains a challenging task for two primary reasons. First, external incentives often drive users’ check-ins, potentially misrepresenting their genuine preferences. Second, while many current research model the temporal dynamics of user preferences in a discrete space, they ignore capturing the continuous evolution of these preferences. To address these challenges, we propose the GraphSAGE-based POI Recommendation via Continuous-Time Modeling (GSA-CTM). We first utilize GraphSAGE to identify real user preferences and filter out noise beyond the user’s real preferences. After GraphSAGE captures complex interaction, we use Gated Recurrent Unit (GRU) combined with neural Ordinary Differential Equations (ODEs) to capture the temporal information embedded in the interaction, and then use neural ODEs to model the user’s continuous dynamic preferences into continuous space. Experiments on two widely-used public datasets validate the superiority of our method.