融合归一化与自适应门控机制的图卷积推荐模型  

Graph Convolutional Recommendation Model IntegratingNormalization and Adaptive Gating Mechanism

作　　者：韩荣 张严 吴贞东[1] 唐宇 张倩[1] 彭子瀚 吴迭 HAN Rong;ZHANG Yan;WU Zhendong;TANG Yu;ZHANG Qian;PENG Zihan;WU Die(College of Computer Science,Sichuan Normal University,Chengdu 610101,Sichuan;College of Leeds,Southwest Jiaotong University,Chengdu 611756,Sichuan)

机构地区：[1]四川师范大学计算机科学学院,四川成都610101 [2]西南交通大学利兹学院,四川成都611756

出　　处：《四川师范大学学报(自然科学版)》2025年第4期534-542,共9页Journal of Sichuan Normal University(Natural Science)

基　　金：国家自然科学基金青年基金(62002250)。

摘　　要：图卷积神经网络广泛应用于推荐任务中,但随着图卷积层数的增加,存在节点特征消失、模型过平滑等问题,导致推荐系统的性能受限.针对上述问题,提出一种融合归一化与自适应门控机制的图卷积推荐模型(FGNGCN).该模型利用图归一化算法,优化节点向量的空间分布,提高节点向量表达能力;引入自适应门控机制,融合原始节点特征向量、图归一化优化向量和图卷积学习向量,解决模型过平滑问题,提高推荐任务的准确率.对3个公开的数据集MovieLens-1M、BeerAdvocate和Yelp进行对比实验,结果表明本模型的评估指标优于现有的多种基线模型.Graph convolutional neural networks are widely used in recommender systems.However,with the increase of the number of graph convolutional layers,there are problems such as node feature disappearance and model oversmoothing,which lead to the performance limitation of recommender systems.Aiming at the above problems,this paper proposes a graph convolution recommendation model(FGNGCN)that integrates normalization and adaptive gating mechanisms.The model uses graph normalization algorithm to optimise the spatial distribution of node vectors to alleviate the influence of the model on recommendation bias.Adaptive gating mechanism is introduced to fuse the original node feature vectors,graph normalization optimisation vectors and graph convolution learning vectors,which solves the over-smoothing problem of the model and improves the accuracy of the recommendation task.Comparative experiments are conducted on three public datasets,MovieLens-1M,BeerAdvocate and Yelp,and the results show that this model outperforms multiple existing baseline models.

关 键 词：推荐系统 图卷积网络 归一化 门控机制 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]