文献类型：期刊文献

中文题名：融合建模的图神经网络会话推荐模型

英文题名：Graph Neural Network Session Recommendation Model with Fusion Modeling

作者：杜佳宇[1];郑红[1];郭津延[1];罗俞建[1];李鹏威[1];单蓉胜[2]

机构：[1]华东理工大学信息科学与工程学院,上海200237;[2]上海交通大学网络空间安全学院,上海200240

年份：2025

卷号：51

期号：6

起止页码：827

中文期刊名：华东理工大学学报(自然科学版)

外文期刊名：Journal of East China University of Science and Technology

收录：;北大核心:【北大核心2023】；

基金：上海市2024年度“科技创新行动计划”资助(24BC3200500,24BC3200300)。

语种：中文

中文关键词：会话推荐;门控图神经网络;图注意力机制;结构化关系;非结构化关系

外文关键词：session recommendation;gated graph neural network;graph attention mechanism;structured relationship;unstructured relationship

摘要：针对传统会话推荐算法仅依赖显式信息而忽视会话间潜在交互关系的问题,本文提出了一种基于门控和图注意力机制的融合建模模型IM-GGN(Integrated Modeling Gated Graph Network),它对物品间的结构化关系和会话间的非结构化关系同时进行建模,从而提升推荐性能。该模型由结构化关系学习(Structured Pattern Learning,SPL)模块与非结构化关系学习(Unstructured Pattern Learning,UPL)模块组成:SPL模块结合图神经网络和门控机制,捕捉会话内部的顺序依赖和长程关系;UPL模块则利用图注意力机制对会话间非结构化的关联信息进行建模,以提取用户偏好上下文。实验结果表明,本文方法在多个公开数据集上均取得了一定程度的性能提升,验证了模型在会话推荐中的有效性。
To address the limitations of traditional session recommendation algorithms that rely solely on explicit information while overlooking potential interactions between sessions,this paper proposes a novel integrated modeling approach based on gating mechanisms and graph attention networks—IM-GGN(Integrated Modeling Gated Graph Network).This model simultaneously captures structured relationships between items and unstructured associations across sessions to enhance recommendation performance.Specifically,the model comprises two main components:the Structured Pattern Learning(SPL)module and the Unstructured Pattern Learning(UPL)module.The SPL module integrates graph neural networks with gating mechanisms to model sequential dependencies and long-range relationships within sessions.Meanwhile,the UPL module leverages graph attention mechanisms to capture unstructured inter-session correlations,thereby extracting contextual user preferences.Experimental results on multiple public datasets demonstrate that the proposed method achieves notable performance improvements,confirming its effectiveness in session-based recommendation tasks.