知识图谱中图结构特征信息学习算法优化  

Optimization of Graph Structure Feature Information Learning Algorithm in Knowledge Graph

作　　者：佘学兵 李祥[2] 明帮铭 SHE Xuebing;LI Xiang;MING Bangming(College of Information Engineering,East China University of Technology,Nanchang 330013;College of Software,East China University of Technology,Nanchang 330013)

机构地区：[1]东华理工大学信息工程学院,南昌330013 [2]东华理工大学软件学院,南昌330013

出　　处：《计算机与数字工程》2022年第8期1755-1759,共5页Computer & Digital Engineering

基　　金：江西省核地学数据科学与系统工程技术研究中心开放基金项目(编号:JETRCNGDSS202001);江西省教育厅重点项目“基于教育云的Docker容器微服务应用架构优化研究”(编号:GJJ170431)资助。

摘　　要：知识图谱采用三元组进行知识表示非常高效,但嵌入到推荐模型中,用户与物品的交互信息、物品的相似度信息被独立建模,忽略了高维的关联特征和长依赖关系,从而导致特征信息提取不充分、推荐效果不佳。为此,提出一种图结构特征信息学习的优化算法。该算法能充分结合知识图谱在推荐中的优势、RNN中的时序特征获取到用户、物品的更深层次的关联特征信息。将模型应用于真实数据集,与现有的基准模型进行对比,实验结果表明该方法能获得更为精准的个性化推荐。The knowledge graph uses triples for knowledge representation,which is very efficient,but embedded into the recommendation model,the interaction information between users and items and the similarity information of items are modeled independently,ignoring the high-dimensional correlation features and long dependencies,resulting in insufficient feature information extraction and poor recommendation effect. Therefore,an optimization algorithm for graph structure feature information learning is proposed. The algorithm can fully combine the advantages of knowledge graph in recommendation and the temporal characteristics in RNN to obtain the deeper correlation feature information of users and items. The model is applied to real data sets and compared with the existing benchmark model. The experimental results show that this method can obtain more accurate personalized recommendation.

关 键 词：图结构 知识图谱 关联特征 个性化推荐 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]