一种知识图谱的排序学习个性化推荐算法  

Personalized Recommendation Algorithm for Learning to Rank by Knowledge Graph

作　　者：杨晋吉[1] 胡波 王欣明[1] 伍昱燊 赵淦森[1] YANG Jin-ji;HU Bo;WANG Xin-ming;WU Yu-shen;ZHAO Gan-sen(School of Computer Science,South China Normal University,Guangzhou 510631,China)

机构地区：[1]华南师范大学计算机学院,广州510631

出　　处：《小型微型计算机系统》2018年第11期2419-2423,共5页Journal of Chinese Computer Systems

基　　金：教育部博士点基金项目(20124407120017)资助;广东省自然科学基金团队项目(S2012030006242)资助

摘　　要：推荐系统是解决"信息过载"的有效方法,提出一种知识图谱的排序学习个性化推荐算法.本文算法首先构建融合上下文信息的知识图谱,使用基于深度学习的网络表示方法 Node2Vec抽取知识图谱特征,通过将排序学习模型产生的反馈模型与用户兴趣迁移模型结合,构建混合推荐模型,最终通过排序学习进行Top-N推荐.该算法能够将各种不同性质的上下文特征结合在一起,并通过排序学习衡量这些多维特征的权重比例,解决了不同特征的融合问题,并且能够考虑到用户兴趣迁移和长短期偏好.在Movielens 1M数据集上的对比实验验证文中算法的有效性,实验表明,该算法能够有效提高推荐的P@N和MAP值.The recommendation system is an effective way to solve the problem of ＂information overload＂.This paper presents a personalized recommendation algorithm for learning to rank by Knowledge Graph.Firstly,the algorithm constructs the Knowledge Graph of the fusion context information.Secondly,we extract the characteristics of the Knowledge Graph use the network representation method based on the Node2Vec.Thirdly,the feedback model generated by the learning to rank model is combined with the user interest migration model to construct the hybrid recommendation model,and finally the Top-N recommendation is done by learning to rank.The algorithm combines the different contextual characteristics of different properties,measures the weight ratio of these multidimensional features by learning to rank.Through this algorithm,we solve the problem of fusion of different features,and take into account the user interest migration and long and short term preferences.The comparison experiment verifies the validity of the algorithm on the Movielens 1M dataset.The experimental results show that this algorithm can effectively improve the P@N value and MAP value of the proposed system.

关 键 词：知识图谱 排序学习 兴趣迁移 Node2Vec 上下文信息 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]