中国科学技术大学学报 ›› 2016, Vol. 46 ›› Issue (1): 82-86.DOI: 10.3969/j.issn.0253-2778.2016.01.011

• 论著 • 上一篇    

基于并行化谱聚类的协同推荐算法研究

郑修猛,陈福才,黄瑞阳   

  1. 国家数字交换系统工程技术研究中心,郑州 450002
  • 收稿日期:2015-08-27 修回日期:2015-09-29 接受日期:2015-09-29 出版日期:2015-09-29 发布日期:2015-09-29
  • 通讯作者: 陈福才
  • 作者简介:郑修猛,男, 1989年生,硕士生. 研究方向:数据挖掘、推荐系统. E-mail:zhengxiumeng@163.com
  • 基金资助:
    国家自然科学基金(61171108),国家重点基础研究发展(973)计划(2012CB315901,2012CB315905),国家科技支撑(863)计划(2014BAH30B01)资助.

Research on collaborative recommendation algorithms based on parallel spectral clustering

ZHENG Xiumeng, CHEN Fucai, HUANG Ruiyang   

  1. China National Digital Switching System Engineering& Technological R&D Center,Zhengzhou 450002,China
  • Received:2015-08-27 Revised:2015-09-29 Accepted:2015-09-29 Online:2015-09-29 Published:2015-09-29

摘要: 随着大规模网络数据的增加,可扩展性成为推荐系统的一个关键因素,为此提出一种基于并行化谱聚类的协同推荐算法.首先通过并行化改进的谱聚类方法对项目进行聚类;然后在基于用户的协同推荐算法基础上,结合已聚类的项目打分信息,提出一种改进的相似用户计算方法,并进行推荐;最后在数据集上进行测试.结果表明,该算法可以有效降低时间复杂度,推荐精确度和推荐效率也有显著提高.

关键词: 推荐系统, 协同过滤, 并行, 谱聚类

Abstract: With the increase of large-scale network data, scalability has become a key factor in the recommendation system. A new collaborative recommendation algorithm is thus based on MapReduce parallel spectral clustering was proposed. First, items are clustered using the improved parallel spectral clustering method; Then, based on the user collaborative recommendation algorithm and combined with the clustered items’ ratings, an improved calculation method for similar users is proposed to establish recommendation. The test results on the dataset show that the proposed algorithm can effectively reduce time complexity, which significantly improving its accuracy and efficiency.

Key words: recommendation system, collaborative filtering, parallel, spectral clustering

中图分类号: