摘要
针对现有会话推荐方法难以从携带噪声的匿名会话序列中精确获取用户偏好的挑战,提出融合对比学习的图神经网络会话推荐模型。首先,将全部会话数据构建为图,通过图神经网络聚合图上物品信息获取节点局部嵌入表示;其次,利用含有噪声滤除器的注意力机制显示地过滤掉不重要的节点表征,得到去噪增强的全局嵌入表示;同时引入对比学习技术设置优化策略指导模型进行去噪学习;最后,采用门控机制为局部和全局表征自适应分配权重,并对其进行加权求和得到用户会话表示,通过预测层生成推荐列表,并在Diginetica和Tmall共2个公开基准数据集上进行实验测试。结果证明所提模型相较于其他基线模型推荐效果明显提升,其中与基于会话推荐的自监督超图卷积网络(S2-DHCN)相比,Tmall上MRR@20提高5.4%,性能显著提高。
Aiming at the problem that existing session recommendation methods are difficult to accurately extract user preferences from anonymous sessions carrying noise,a graph neural network model for session-based recommendation is presented in this paper.Specifically,this method first constructed all the session data as a graph,obtaining the local embedded representation of the node by aggregating the item information on the graph through the graph neural network.Secondly,the attention mechanism of the noise filter was used to visually filter out the unimportant node representations to obtain the global embedding representation of denoising enhancement,and the comparative learning technology was introduced to set the optimization strategy to guide the model for denoising learning.Finally,the gating mechanism was used to adaptively assign weights to local and global representations,the weighted summation was used to obtain the user session representation,and then the recommendation list was generated through the prediction layer.Experimental tests were carried out on two public benchmark datasets of Diginetica and Tmall.The results show that the recommendation effect of the proposed model is better than other baseline methods,and the MRR@20 on Tmall is improved by 5.4%compared with the self-supervised hypergraph convolutional network based on session recommendation(S2-DHCN),and the recommendation performance is significantly improved.
作者
郑小丽
王巍
张闯
杜雨晅
ZHENG Xiaoli;WANG Wei;ZHANG Chuang;DU Yuxuan(Hebei Key Laboratory of Security&Protection Infor-mation Sensing&Processing,Hebei University of Engineering,Handan 056038,China;College of Internet of Things Engineering,Jiangnan University,Wuxi 214122,China)
出处
《应用科技》
CAS
2023年第5期8-16,共9页
Applied Science and Technology
基金
国家自然科学基金项目(61802107)
河北省高等学校科学技术研究项目(ZD2020171)
江苏省博士后科研资助项目(1601085C).
关键词
会话推荐
图神经网络
注意力机制
对比学习
噪声滤除器
去噪增强
会话表示
用户偏好
session sequence recommendation
graph neural network
attention mechanism
contrasting learning
noise filter
denoising enhancement
session representation
user preference
