摘要
针对现有模型对异质信息网络(heterogeneous information network, HIN)信息提取大部分依赖于元路径,缺乏元路径信息补充以及很少学习异质图中复杂的结构信息等问题,提出一种异质网中基于邻居节点和元路径的推荐算法(NMRec)。提取用户和物品邻居节点补充元路径缺失的信息,以卷积的方式捕获节点之间丰富的交互,通过注意力机制得到节点和元路径的嵌入表示,拼接用户、物品、邻居节点及元路径进行TOP-N推荐。在两个公开数据集上的实验结果表明,NMRec推荐性能良好,对推荐结果有良好的可解释性,与7种推荐基准算法相比,NMRec在评价指标Pre@10、Recall@10、NDGG@10上至少提升了0.21%、29%、1.46%。
As for existing models,heterogeneous information network(HIN)information extraction mostly rely on meta-paths,lack meta-path information supplement and rarely learn complex structure information in heterogeneous maps,etc.,a recommendation algorithm(NMRec)based on neighbor nodes and meta paths in heterogeneous networks was proposed.The information of missing metapathes was extracted from users and item neighbor nodes.Rich interactions between nodes were captured in the way of convolution,and the embedded representation of nodes and metapathes was obtained through the attention mechanism.Users,items,neighbor nodes and metapathes were spliced together for TOP-N recommendation.Experimental results on two public data sets show that NMRec has good recommendation performance and interpretability for recommendation results.Compared with seven recommendation benchmark algorithms,NMRec least improves the evaluation indexes Pre@10,Recall@10 and NDGG@10 by 0.21%,29%and 1.46%.
作者
贵向泉
张榕榕
李立
GUI Xiang-quan;ZHANG Rong-rong;LI Li(School of Computer and Communication,Lanzhou University of Technology,Lanzhou 730050,China)
出处
《计算机工程与设计》
北大核心
2024年第7期2050-2056,共7页
Computer Engineering and Design
基金
国家重点研发计划基金项目(2020YFB1713600)
甘肃省重点研发计划-工业类基金项目(22YF7GA159)
甘肃省基础研究计划-软科学专项基金项目(22JR4ZA084)。
关键词
异质信息网络
表示学习
元路径
邻居信息
注意力机制
卷积神经网络
推荐系统
heterogeneous information network
presentation learning
meta-path
neighbor information
attention mechanism
convolutional neural network
recommendation system
