融合图注意力网络和注意力因子分解机的服务推荐方法  

Service recommending method incorporating graph attentionnetwork and attention factorization machine

作　　者：黄德玲[1] 童夏龙 杨皓栋 HUANG Deling;TONG Xialong;YANG Haodong(School of Software Engineering,Chongqing University of Posts and Telecommunications,Chongqing 400065,P.R.China)

机构地区：[1]重庆邮电大学软件工程学院,重庆400065

出　　处：《重庆邮电大学学报（自然科学版）》2024年第2期357-366,共10页Journal of Chongqing University of Posts and Telecommunications(Natural Science Edition)

基　　金：重庆市技术创新与应用发展专项重点项目(cstc2021jscx-gksbX0027)。

摘　　要：为了解决服务推荐过程中的特征稀疏问题、提高服务的语义表示能力,进而提升推荐的准确性和有效性,提出基于图注意力网络(graph attention networks,GAT)研究服务推荐方法,引入服务的组合权重和组合的结构信息,综合利用多种服务特征,提高服务推荐质量。将GAT和注意力因子分解机(attention factorization machine,AFM)结合在一起,利用多头自注意力机制,学习每个节点在图邻域中的重要性;进行信息聚合,处理网络中的不同图结构,以适应服务动态变化的场景。实验结果显示,在数据平衡的情况下,提出的方法性能表现均好于对比方法;在数据不平衡的情况下,提出的方法大部分性能指标也表现良好,达到了提升服务推荐准确性和有效性的目标。To address feature sparsity in service recommendation and enhance semantic representation,we introduce a method leveraging graph attention networks(GAT).This approach integrates service combination weights and structural information,leveraging multiple service features for enhanced recommendation accuracy.By combining GAT with attention factorization machine(AFM)and employing a multi-head self-attention mechanism,the model learns node importance within the graph neighborhood.This enables effective information aggregation across diverse graph structures,adapting to dynamic service changes.Experimental results demonstrate the superiority of the proposed method compared to alternative approaches,particularly in balanced data scenarios.Moreover,most performance metrics remain strong even in imbalanced data settings,showcasing the method’s efficacy in improving recommendation accuracy and effectiveness.

关 键 词：服务推荐 图注意力网络 注意力因子分解机 应用程序接口 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]