摘要
当前的推荐方法普遍引入知识图谱作为辅助信息来缓解协同过滤算法的缺陷,如数据稀疏和冷启动问题.然而,基于知识图谱的推荐方法大都专注于利用知识图谱来构建用户及物品的特征表示,忽略了对用户交互信息的有效利用.本文在引入用户-物品交互图和知识图谱两种图结构信息基础上,通过图神经网络融合用户-物品间的交互特征、物品间的相似特征以及知识图谱中实体的知识特征,来构建用户物品的特征表示,并将之应用于推荐系统.实验表明,相对于基线模型,本文提出的模型具有较好的推荐效果.
Current recommendation methods generally introduce knowledge graph(KG)as auxiliary information to alleviate the defects of collaborative filtering algorithm,such as data sparsity and cold start.However,these KG-aware recommendation methods solely focus on constructing feature representation of users and items through KG,and ignore the effective utilization of user interactive information.On the basis of introducing two kinds of graph-structure data,user-item interaction graph and KG,this paper applies GNN to integrate the interactive feature between users and items,the similar feature between items and the knowledge feature in entities from KG,to construct the feature representation of users and items,and apply it to recommender system.Experiments show that compared with the baseline models,our proposed model has a better recommendation effect.
作者
杨中金
彭敦陆
宋祎昕
YANG Zhongjin;PENG Dunlu;SONG Yixin(School of Optical-Electrical and Computer Engineering,University of Shanghai for Science and Technology,Shanghai 200093,China;National Pipeline Network Group Zhejiang Natural Gas Pipeline Network Co.,Ltd.,Hangzhou 310051,China)
出处
《小型微型计算机系统》
CSCD
北大核心
2024年第8期1895-1900,共6页
Journal of Chinese Computer Systems
基金
国家自然科学基金项目(61772342)资助.
关键词
协同过滤
推荐系统
用户-物品交互图
知识图谱
collaborative filtering
recommender system
user-item interaction graph
knowledge graph
