计算机应用

• 典型应用(Typical applications) • 上一篇    下一篇

基于HMM模型的协同过滤推荐方法

黄光球 赵永梅   

  1. 西安建筑科技大学 管理学院 西安建筑科技大学 管理学院
  • 收稿日期:2007-12-14 修回日期:2008-01-28 发布日期:2008-06-01 出版日期:2008-06-01
  • 通讯作者: 黄光球

Approach to collaborative filtering recommendation based on HMM

Guang-qiu HUANG Yong-mei ZHAO   

  • Received:2007-12-14 Revised:2008-01-28 Online:2008-06-01 Published:2008-06-01
  • Contact: Guang-qiu HUANG

摘要: 考虑到用户浏览路径、时间、浏览次数都是影响推荐准确度的重要因素,提出一种基于隐马尔可夫模型(HMM)的动态协同过滤推荐方法。该方法首先用HMM模型模拟用户浏览网页时的行为,根据用户浏览网页时的行为建立最近邻集合。由于数据不是简单的用户评分,而是用户浏览网页的路径,这样就解决了数据稀疏问题和最初评价问题。并且使用HMM代替简单的相似模型来度量用户相似性,提高了最近邻推荐的准确性,解决了实时性推荐和数据空间的可扩展的问题。然后,提出了喜好度的概念并给出了计算方法,喜好度概念的加入能为目标用户推荐更适合的商品。最后,结合喜好度给出了基于HMM的协同过滤预测模型。通过对一个实例的研究验证了所提出的算法以及推荐模型的可行性。

关键词: 隐马尔可夫模型, 协同过滤推荐, 前(后)向评估算法, 浏览路径, 喜好度

Abstract: Considering that browse route, browse time, browse times and so on are the important factors to influence the accuracy of commendation, a dynamic collaboration filtering recommendation method based on Hidden Markov Model (HMM) was proposed. First, it simulated users' behaviors while a user was browsing Web pages, and set up the nearest-neighbor set according to his behaviors. Because the data it used was not users' rating but users' browse route, the problem of data sparseness and initial rating was resolved. When HMM was used to replace the similitude model to measure users' similarity, the accuracy of nearest-neighbor commendation was improved greatly. And it settled the on-time recommendation problem and the extensible data space problem. Then the concept of fancy degree was set up, which made the recommendation become more suitable. Finally, the fancy degree was applied to establish the prediction model of dynamic collaboration filtering recommendation. A case study shows the excellent performance of this model.

Key words: Hidden Markov Model (HMM), collaborative filtering recommendation, forward (backward) estimation algorithm, browse route, fancy degree