Multifaceted Scenario-Aware Hypergraph Learning for Next POI Recommendation

TL;DR

MSAHG tackles multi-scenario next POI recommendation by explicitly modeling context-specific mobility through scenario-specific multi-view sub-hypergraphs and by mitigating cross-scenario interference with an adaptive parameter-splitting mechanism. It constructs eight sub-hypergraphs across collaborative, temporal, geographical, and transitional views, applies two-step hypergraph convolution with residuals, and fuses multi-view embeddings through gating while enforcing cross-view consistency via contrastive learning. The adaptive splitting duplicates conflicting parameters to separate scenario updates, enabling robust joint learning. Empirical results on Gowalla, NYC, and TKY show consistent gains over five SOTA baselines across diverse scenario settings, along with thorough ablations and analyses of category and distance distributions, indicating strong practical impact for multi-scenario POI recommendation.

Abstract

Among the diverse services provided by Location-Based Social Networks (LBSNs), Next Point-of-Interest (POI) recommendation plays a crucial role in inferring user preferences from historical check-in trajectories. However, existing sequential and graph-based methods frequently neglect significant mobility variations across distinct contextual scenarios (e.g., tourists versus locals). This oversight results in suboptimal performance due to two fundamental limitations: the inability to capture scenario-specific features and the failure to resolve inherent inter-scenario conflicts. To overcome these limitations, we propose the Multifaceted Scenario-Aware Hypergraph Learning method (MSAHG), a framework that adopts a scenario-splitting paradigm for next POI recommendation. Our main contributions are: (1) Construction of scenario-specific, multi-view disentangled sub-hypergraphs to capture distinct mobility patterns; (2) A parameter-splitting mechanism to adaptively resolve conflicting optimization directions across scenarios while preserving generalization capability. Extensive experiments on three real-world datasets demonstrate that MSAHG consistently outperforms five state-of-the-art methods across diverse scenarios, confirming its effectiveness in multi-scenario POI recommendation.

Figures (10)

• Figure 1: Illustration of scenario-specific user preferences across three contextual dimensions in LBSNs, highlighting substantial behavioral differences across scenarios.
• Figure 2: POI category distributions in NYC trajectories of local users and tourists. The marked differences, especially in Hotel, Residence, and Transportation categories, underline the necessity of scenario-aware modeling.
• Figure 3: The overall framework of Multifaceted Scenario-Aware Hypergraph Learning (MSAHG). User trajectories are categorized into three scenario dimensions: user type, temporal context, and spatial region. Based on these and relational patterns, eight sub-hypergraphs are constructed and processed via multi-layer hypergraph convolution. A parameter splitting mechanism handles conflicting scenario signals. Final embeddings are learned through contrastive learning and fused for next POI prediction.
• Figure 4: Overview of adaptive parameter splitting. For each parameter $\theta$, gradients from different scenarios are compared. When conflicting directions are detected, $\theta$is duplicated, and scenarios are grouped to use either $\theta$ or the new copy $\theta'$.
• Figure 5: Comparison of category $\Delta$Percentage (difference between predicted and true category proportions) for DCHL and MSAHG in Local and Tourist scenarios on the NYC dataset. MSAHG better preserves the original distribution.