Graph-Sequential Alignment and Uniformity: Toward Enhanced Recommendation Systems
Yuwei Cao, Liangwei Yang, Zhiwei Liu, Yuqing Liu, Chen Wang, Yueqing Liang, Hao Peng, Philip S. Yu
TL;DR
Addresses the limitation of using either graph-based or sequential recommenders in isolation by proposing GSAU, a shared-embedding framework that jointly trains GNN and sequential encoders. The approach preserves the original architectures and introduces a cross-encoder loss that enforces alignment and uniformity across and within encoders, via losses $L_A$, $L_U$, $L_{GA}$, $L_{GU}$, $L_{SA}$, $L_{SU}$ and a balancing weight $\\gamma$; the overall objective is $\\mathcal{L} = \\mathcal{L}_{GA} + \\mathcal{L}_{SA} + \\gamma (\\mathcal{L}_{GU} + \\mathcal{L}_{SU})$. Experiments on three real-world datasets demonstrate substantial improvements over dedicated graph or sequential models, with GSAU (rec) achieving state-of-the-art results and a best $\\gamma$ around 0.1. The code is publicly available at https://github.com/YuweiCao-UIC/GSAU.git.
Abstract
Graph-based and sequential methods are two popular recommendation paradigms, each excelling in its domain but lacking the ability to leverage signals from the other. To address this, we propose a novel method that integrates both approaches for enhanced performance. Our framework uses Graph Neural Network (GNN)-based and sequential recommenders as separate submodules while sharing a unified embedding space optimized jointly. To enable positive knowledge transfer, we design a loss function that enforces alignment and uniformity both within and across submodules. Experiments on three real-world datasets demonstrate that the proposed method significantly outperforms using either approach alone and achieves state-of-the-art results. Our implementations are publicly available at https://github.com/YuweiCao-UIC/GSAU.git.