摘要： 针对现有推荐系统仅考虑用户兴趣偏移的随机性,而忽略用户兴趣偏移时效性的问题,通过研究马尔可夫模型,并引入滑动时间窗口机制,提出一种新的多阶马尔可夫预测推荐算法。该算法通过学习用户历史行为数据,以及分析用户浏览行为特征,达到准确预测用户浏 览行为的目的。实验结果表明,与协同过滤算法相比,该推荐算法不仅能够针对用户兴趣的偏移进行有效预测推荐,而且运行时间较短。

关键词: 用户兴趣, 马尔可夫模型, 随机性, 时效性, 滑动时间窗口, 推荐系统

Abstract: The current research on the recommendation system just considers the randomness too much but ignores the point of timeliness.This paper proposes a new multi-order Markov predicting recommendation system based on the traditional standard Markov model.The algorithm can make up for the deficiency of timeliness by using the new sliding time window mechanism.The algorithm can achieve the mission of accurate prediction about user’s browsing activity according to the study of the history data and the analysis of user’s browsing habits.Experimental results show that compared with the collaborative filtering scheme,the new proposed algorithm can provide the predicting recommendation efficiently and effectively judging by the user’s interest shift behavior.

Key words: user interest, Markov model, randomness, timeliness, sliding time window, recommendation system

中图分类号: 

• TP391