基于学习排序的并行协同过滤推荐算法  

Parallel collaborative filtering recommender algorithm based on learning to rank

作　　者：肖菁[1] 袁凌[3] 黄昌勤[2] 吴不晓[1] Xiao Jing1a, Yuan Ling2 ,Huang Changqin1b, Wu Buxiao1a(1 a School of Computer Science, b School of Information Technology in Education, South China Normal University, Guangzhou 510631, China; 2 School of Computer Science and Technology, Huazhong University of Science and Technology, Wuhan430074, Chin)

机构地区：[1]华南师范大学计算机学院,广东广州510631 [2]华南师范大学教育信息技术学院,广东广州510631 [3]华中科技大学计算机科学与技术学院,湖北武汉430074

出　　处：《华中科技大学学报（自然科学版）》2018年第3期36-41,共6页Journal of Huazhong University of Science and Technology(Natural Science Edition)

基　　金：国家自然科学基金资助项目(61202296,61370178);广东省科技计划专项资助项目(2015A030401087);广州市科技计划资助项目(201604016019,201604010003)

摘　　要：为实现大数据环境下高效、精准的商品推荐,将协同过滤思想与信息检索理论有机融合,提出基于学习排序（LTR）的并行协同过滤推荐算法.首先利用相似物品-物品网络图结构共享参数的方法减少参数,通过Pairwise方法构造目标函数,并利用梯度上升法得到参数.还提出了通过使用层次聚类的方法对差异性较大的相似图进行分裂,以保证推荐的准确度.最后给出大数据平台Spark下该推荐算法的并行化实现方案.在真实数据集Netflix上的实验结果表明：提出的算法不仅在召回率和准确率上有所提高,而且计算效率高效,表明该方法可以应用于大数据场景中的推荐服务.In order to provide efficient and accurate item recommendation in big data case,a collaborative filtering recommendation algorithm based on learning to rank（LTR） which combined the collaborative filtering theory and the method in information retrieval was proposed.The method of sharing parameters among similar item-item network graph structures was used firstly to reduce the learning parameter numbers.The object function was constructed using Pairwise method and the learning parameters were obtained by gradient ascent method.A splitting method was introduced accordingly through hierarchy clustering to divide a network graph structure into smaller ones.The parallelization details for Spark platform was given accordingly.Experimental results of real dataset Netflix show that the proposed algorithm performs better in terms of precision and recall,and it can be used for recommendation services in big data environment.

关 键 词：推荐算法 协同过滤 学习排序(LTR) TOP N推荐 并行化 Spark平台 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]