• 中国计算机学会会刊
  • 中国科技核心期刊
  • 中文核心期刊

计算机工程与科学

• 论文 • 上一篇    下一篇

基于时间效应的协同过滤算法

吴飞,余腊生,冯梅   

  1. (中南大学软件学院,湖南 长沙 410083)
  • 收稿日期:2015-12-25 修回日期:2016-06-07 出版日期:2017-11-25 发布日期:2017-11-25
  • 基金资助:

    国家自然科学基金(61573380)

A collaborative filtering algorithm based on time effect

WU Fei,YU La-sheng,FENG Mei   

  1. (School of Software,Central South University,Changsha 410083,China)
  • Received:2015-12-25 Revised:2016-06-07 Online:2017-11-25 Published:2017-11-25

摘要:

协同过滤算法已被成功应用在个性化推荐系统中,但传统的协同过滤算法很少考虑时间因素的影响,难以确保最近邻集的准确性和可靠性。虽然很多文献提出了各种改进推荐算法,但仍然没能在计算中有效地将时间因素和用户评分综合起来。因此,在原有的工作基础上提出基于时间效应的协同过滤算法,将时间因素纳入用户预测评分和用户相似性计算中,并综合这两个因素来动态分配每一项评分的权重,采用预测评分填充用户-项矩阵和二次计算用户相似性矩阵的方法,最终得到Top-N推荐集。实验表明,改进后的算法提高了推荐算法的精度和推荐质量。

 

关键词: 协同过滤, 时间效应, 项集相似性, 预测评分, 平均绝对误差

Abstract:

Collaborative filtering algorithms have been successfully applied in personalized recommendation. However, it is hard for traditional filtering algorithms to make sure that the accuracy and reliability of the nearest neighbors set is good enough because they ignore the impact of time factor. Even though there are a number of improved collaborative filtering algorithms, they cannot integrate time factor and user scores in their calculation. We propose a new algorithm based on time factor and our original work, which introduces time factor into user score prediction and user similarity calculation and assigns each item a dynamic weight by synthesizing time and user similarity. We finally obtain a Top-N set by filling the user-item matrix with user prediction scores and secondary calculating user similarity matrix. Experiments prove that the improved algorithm can enhance  recommendation accuracy and quality.
 

Key words: collaborative filtering, time effect, items set similarity, predict score, mean absolute error