Bi-Level Graph Structure Learning for Next POI Recommendation

TL;DR

A novel bi-level structure learning scheme that employs a multi-relational graph network that considers both POI- and prototype-level neighbors, resulting in improved POI representations and improves the exploration ability for recommendation by alleviating sparsity issues.

Abstract

Next point-of-interest (POI) recommendation aims to predict a user's next destination based on sequential check-in history and a set of POI candidates. Graph neural networks (GNNs) have demonstrated a remarkable capability in this endeavor by exploiting the extensive global collaborative signals present among POIs. However, most of the existing graph-based approaches construct graph structures based on pre-defined heuristics, failing to consider inherent hierarchical structures of POI features such as geographical locations and visiting peaks, or suffering from noisy and incomplete structures in graphs. To address the aforementioned issues, this paper presents a novel Bi-level Graph Structure Learning (BiGSL) for next POI recommendation. BiGSL first learns a hierarchical graph structure to capture the fine-to-coarse connectivity between POIs and prototypes, and then uses a pairwise learning module to dynamically infer relationships between POI pairs and prototype pairs. Based on the learned bi-level graphs, our model then employs a multi-relational graph network that considers both POI- and prototype-level neighbors, resulting in improved POI representations. Our bi-level structure learning scheme is more robust to data noise and incompleteness, and improves the exploration ability for recommendation by alleviating sparsity issues. Experimental results on three real-world datasets demonstrate the superiority of our model over existing state-of-the-art methods, with a significant improvement in recommendation accuracy and exploration performance.

Figures (7)

• Figure 1: Hierarchical structures in different POI features, i.e., fine-grained POIs can be divided into coarse-grained groups (corresponding to cyan and red triangles in the figure). We use prototypes to represent coarse-grained group information and introduce prototype nodes in the graph structure learning to construct hierarchical graphs.
• Figure 2: The overall framework of our proposed BiGSL model. We first construct multiple feature views from primitive POI features. In each view, we map POIs to nodes and then cluster the POI features to reveal the hierarchical structure. The resulting prototypes represent the coarse-grained group information, which are added to the graph. Subsequently, we perform pairwise structure learning to infer the connectivity between POI pairs and prototype pairs, resulting in a data-driven hierarchical graph. Based on the hierarchical graph, we conduct the multi-relational graph learning to produce better POI representations. Finally, to encourage POI information fusion from different views and make better recommendations, we adopt a contrastive multiview fusion approach by mining view-shared and view-specific information.
• Figure 3: The neighborhood of target node "" for graph representation learning. We define three types of neighbor nodes: "" denotes the POI-level neighbor nodes, "" and "" denote the 1-hop and 2-hop prototype-level neighbor nodes, respectively.
• Figure 4: The t-SNE visualization of learned structure embedding on the Gowalla dataset. Colors indicate clustering labels. The introduction of $\mathcal{L}_{\text{HSL}}$ can help the model to effectively capture the clustering and hierarchical information in POI features, and the clustering property of spatial features is more pronounced than that of temporal features.
• Figure 5: Effect of different cluster numbers and loss weights. The y-axis represents accuracy and the x-axis is the different hyper-parameter values.