Open Access

ARTICLE

HMGS: Hierarchical Matching Graph Neural Network for Session-Based Recommendation

Pengfei Zhang¹, Rui Xin¹, Xing Xu¹, Yuzhen Wang¹, Xiaodong Li², Xiao Zhang², Meina Song², Zhonghong Ou^3,*

1 State Grid Hebei Information and Telecommunication Branch, Shijiazhuang, 050000, China
2 School of Computer Science (National Pilot Software Engineering School), Beijing University of Posts and Telecommunications, Beijing, 100876, China
3 State Key Laboratory of Networking and Switching Technology, Beijing University of Posts and Telecommunications, Beijing, 100876, China

* Corresponding Author: Zhonghong Ou. Email:

Computers, Materials & Continua 2025, 83(3), 5413-5428. https://doi.org/10.32604/cmc.2025.062618

Received 23 December 2024; Accepted 14 March 2025; Issue published 19 May 2025

Abstract

Session-based recommendation systems (SBR) are pivotal in suggesting items by analyzing anonymized sequences of user interactions. Traditional methods, while competent, often fall short in two critical areas: they fail to address potential inter-session item transitions, which are behavioral dependencies that extend beyond individual session boundaries, and they rely on monolithic item aggregation to construct session representations. This approach does not capture the multi-scale and heterogeneous nature of user intent, leading to a decrease in modeling accuracy. To overcome these limitations, a novel approach called HMGS has been introduced. This system incorporates dual graph architectures to enhance the recommendation process. A global transition graph captures latent cross-session item dependencies, while a heterogeneous intra-session graph encodes multi-scale item embeddings through localized feature propagation. Additionally, a multi-tier graph matching mechanism aligns user preference signals across different granularities, significantly improving interest localization accuracy. Empirical validation on benchmark datasets (Tmall and Diginetica) confirms HMGS’s efficacy against state-of-the-art baselines. Quantitative analysis reveals performance gains of 20.54% and 12.63% in Precision@10 on Tmall and Diginetica, respectively. Consistent improvements are observed across auxiliary metrics, with MRR@10, Precision@20, and MRR@20 exhibiting enhancements between 4.00% and 21.36%, underscoring the framework’s robustness in multi-faceted recommendation scenarios.

---

Keywords

---