计算机工程与应用 ›› 2015, Vol. 51 ›› Issue (23): 119-124.

• 数据库、数据挖掘、机器学习 • 上一篇    下一篇

一种优化标签的矩阵分解推荐算法

张  明1,2,郭  娣3   

  1. 1.武汉理工大学 信息工程学院,武汉 430070
    2.临沂大学 信息学院,山东 临沂 276000
    3.上海理工大学 光电信息与计算机工程学院,上海 200093
  • 出版日期:2015-12-01 发布日期:2015-12-14

Matrix factorization recommendation algorithm of optimizing tags

ZHANG Ming1,2, GUO Di3   

  1. 1.School of Information Engineering, Wuhan University of Technology, Wuhan 430070, China
    2.School of Information, Linyi University, Linyi, Shandong 276000, China
    3.School of Optical-Electrical and Computer Engineering, University of Shanghai for Science and Technology, Shanghai 200093, China
  • Online:2015-12-01 Published:2015-12-14

摘要: 个性化推荐研究中,垃圾标签不仅会导致数据稀疏性问题,同时影响推荐的实时性和精确性。因此提出一种优化标签的矩阵分解推荐算法OTMFR,该算法分为两个阶段:首先优化标签,在建立三部网络图的基础上提出一种标签排序算法,利用互增强的关系得到关于标签流行度的排序,去除排序靠后的垃圾标签;然后在此基础上利用用户和资源对标签的偏好信息构建用户-资源偏好矩阵,并从矩阵分解的角度为用户产生推荐。在Delicious数据集上的实验结果表明,该算法在推荐精准度上有较为明显的效果。

关键词: 标签, 网络图, 互增强, 偏好信息, 矩阵分解

Abstract: In personalized recommendation, garbage tags not only lead to the problem of data sparsity, but also affect both timeliness and accuracy of recommendation. Therefore, this paper presents a matrix factorization recommendation algorithm of optimizing tags OTMFR, which is divided into two stages:first optimizing tags, a tags sorting algorithm is proposed based on establishing tripartite graphs network, which gets tags popularity ranking by making use of enhanced mutual relations and removes garbage tags those rank rearward. Then it uses users’ and resources’ preference information on tags to build the user-resource preference matrix, and generates recommendations for users from the perspective of matrix factorization. The experimental results on the dataset of Delicious show that the algorithm has better recommendation results.

Key words: tags, graphs network, enhanced mutual relations, preference information, matrix factorization