Distance-aware Self-adaptive Graph Convolution for Fine-grained Hierarchical Recommendation

TL;DR

This work tackles two core issues in graph convolutional recommender systems: optimizing the representation space and properly weighing multi-layer information during aggregation. It introduces SAGCN, a distance-based adaptive hierarchical aggregation framework that learns hierarchical embedding weights from inter-layer distances using a distance set including $dist_E$, $dist_C$, and $dist_K$, and a weighting scheme with $score_{old}$, $score_{new}$, $w_{old}$, and $w_{new}$. The model integrates a three-part architecture (graph representation, hierarchical aggregation, and prediction) and optimizes with a BPR loss, evaluated on four public datasets where it achieves notable gains over strong baselines like LightGCN. The results demonstrate improved cross-layer information fusion, enhanced embedding space properties, and practical improvements in recall and ranking metrics, suggesting strong potential for broader applicability in graph-based recommendation tasks.

Abstract

Graph Convolutional Networks (GCNs) are widely used to improve recommendation accuracy and performance by effectively learning the representations of user and item nodes. However, two major challenges remain: (1) the lack of further optimization in the graph representation structure and (2) insufficient attention given to the varying contributions of different convolutional layers.This paper proposes SAGCN, a distance-based adaptive hierarchical aggregation method that refines the aggregation process through differentiated representation metrics. SAGCN introduces a detailed approach to multilayer information aggregation and representation space optimization, enabling the model to learn hierarchical embedding weights based on the distance between hierarchical representations. This innovation allows for more precise cross-layer information aggregation, improves the model's ability to capture hierarchical embeddings, and optimizes the representation space structure. Additionally, the objective loss function is refined to better align with recommendation tasks.Extensive experiments conducted on four real-world datasets demonstrate significant improvements, including over a 5% increase on Yelp and a 5.58% increase in Recall@10 on the ML_1M dataset.

Figures (6)

• Figure 1: Overall framework of SAGCN
• Figure 2: Comparison of Recall with benchmark model performance
• Figure 3: Comparison of NDCG with benchmark model performance
• Figure 4: Comparison of recall Performance of Different Distance Metrics with Benchmark Model
• Figure 5: Comparison of NDCG Performance of Different Distance Metrics with Benchmark Model