基于混合隐Markov链浏览模型的WEB用户聚类与个性化推荐

Web User Clustering and Personalized Recommendation Based on Mixtures of Hidden Markov Chain Models

针对传统的Markov链模型不能有效的表征长串访问序列所蕴含的丰富的用户行为特征（用户类别特征、访问兴趣迁移特征）的缺点,提出混合隐Markov链浏览模型。混合隐Markov链模型使用多个不同的模型来区分不同类别用户的浏览特征,并为每个类别的用户设置了能跟踪捕捉其访问兴趣变化的类隐Markov链模型,能更好地对WWW长串访问序列的复杂特征进行建模,在真实WWW站点访问日志数据上的用户聚类实验与个性化推荐实验的结果表明,混合隐Markov链模型与传统的Markov链模型相比,具有更理想的聚类性能和推荐性能。

Since the Markov Chain Model can not denote the abundant users ＇ behavioral characteristics（such as： characteristics of users＇ type,characteristics of users＇ interests transfer） of a long access sequence effectively,the Mixtures of Hidden Markov Chain Models is proposed.Mixtures of Hidden Markov Chain Models use different models to distinguish the browsing categories of users from different types,and set a Hidden Markov Chain Models（can track and catch the changes of users＇ interests） for each users＇ type.Mixtures of Hidden Markov Chain Models can model the complex characteristics of the WWW long access sequences better.The results of users clustering experiment and personalized recommendation experiment with a real WWW web access log data show that Mixtures of Hidden Markov Chain Models have more perfect clustering and recommendation performance than Markov Chain Model.