基于对偶四元数的协同知识图谱推荐模型  

Dual Quaternion Based Collaborative Knowledge Graph Modeling for Recommendation

作　　者：曹宗胜 许倩倩[2] 李朝鹏 姜阳邦彦 操晓春[1,4] 黄庆明 CAO Zong-Sheng;XU Qian-Qian;LI Zhao-Peng;JIANG Yangbangyan;CAO Xiao-Chun;HUANG Qing-Ming(State Key Laboratory of Information Security,Institute of Information Engineering,Chinese Academy of Sciences,Beijing 100093;Key Laboratory of Intelligent Information Processing,Institute of Computing Technology,Chinese Academy of Sciences,Beijing 100190;School of Computer Science and Technology,University of Chinese Academy of Sciences,Beijing 101408;School of Cyber Security,University of Chinese Academy of Sciences,Beijing 100049;Key Laboratory of Big Data Mining and Knowledge Management(BDKM),Chinese Academy of Sciences,Beijing 101408;Peng Cheng Laboratory,Shenzhen,Guangdong 518055)

机构地区：[1]中国科学院信息工程研究所信息安全国家重点实验室,北京100093 [2]中国科学院计算技术研究所智能信息处理重点实验室,北京100190 [3]中国科学院大学计算机科学与技术学院,北京101408 [4]中国科学院大学网络空间安全学院,北京100049 [5]中国科学院大数据挖掘与知识管理重点实验室,北京101408 [6]鹏城实验室,广东深圳518055

出　　处：《计算机学报》2022年第10期2221-2242,共22页Chinese Journal of Computers

基　　金：科技创新2030-“新一代人工智能”重大项目(2018AAA0102000);国家自然科学基金项目(U21B2038,61931008,6212200758,61976202);中央高校基本科研业务费专项资金、中科院青促会会员项目、中国科学院战略性先导科技专项(XDB28000000)资助.

摘　　要：为减轻用户-商品交互数据稀疏和冷启动问题对协同过滤算法推荐效果的影响,许多研究者将知识图谱引入推荐系统.然而既往工作对协同知识图谱中关系的建模较为单一,表达能力较弱,不利于建模实体间的复杂关系以及学习用户和商品的嵌入.为克服这一问题,本文提出基于对偶四元数的协同知识图谱推荐模型(DQKGR),利用富有表达力的对偶四元数嵌入表示用户和商品,以有效建模实体和关系之间复杂的潜在依赖.在商品和用户的嵌入学习部分,本文设计了一个新的知识图谱嵌入模型(DQKGE),可基于莫比乌斯变换捕捉协同知识图谱中实体间多种复杂关系,并基于此设计偏好传播与聚合方法以利用知识图谱中的结构信息进行推荐.为验证所提DQKGR模型的有效性,在公开的Last-FM、MovieLens-20M和Book-Crossing数据集上进行大量实验,结果表明所提DQKGR模型推荐结果在多个评测指标上优于现有方法,可在平均意义上产生2.83%以上的性能提升.To alleviate the issues caused by sparse user-item interaction data and cold-start problems,which limit the recommendation performance of collaborative filtering,researchers integrate the knowledge graph(KG)to recommender systems.However,most existing KG-aware recommendation models use the single transformational method for KG embedding,which may limit the modeling of complicated relations existing in real-world data.Besides,these models often represent users and items using real-valued embeddings,which are of less representation capacity.In this paper,we propose Dual Quaternion-based Collaborative Knowledge Graph Modeling for Recommendation(DQKGR),which represents users and items with dual quaternion embeddings in hypercomplex space,so that the latent inter-dependencies between entities and relations could be captured effectively.In the core of our model,we propose a method called Dual Quaternion-based and Mobius Transformation-based Knowledge Graph Embeddings(DQKGE),which can capture multiple complex relations in collaborative KGs.On top of this,those embeddings are updated by a customized preference propagation and aggregation method with structure information concerned.Finally,we apply the proposed DQKGR to three real-world datasets,including Last-FM,MovieLens-20M,and Book-Crossing.Results show that DQKGR outperforms existing methods on several metrics,and achieves a more than 2.83%performance gain in average.

关 键 词：知识图谱 推荐系统 对偶四元数 莫比乌斯变换 偏好传播 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]