基于会话聚类和马尔科夫链的动态用户行为模型改进研究

Improvement Research of Dynamic User Behavior Model Based on Session Clustering and Markov Chain

摘要: 根据点击流建立用户行为模型并对用户特征进行分析，是企业制定精准的营销策略及提供个性化推荐的基础。首先，本文在传统的动态用户行为模型(CBMG)基础上提出了一种同时考虑页面类型和行为序列的改进的用户行为模型，以从用户路径偏好信息中充分反映其行为模式。其次，基于行为序列和页面类型对用户会话进行聚类，得到的不同行为模式的会话类别，针对不同会话类型基于马尔科夫链得到用户行为状态转移的动态模型。研究发现：基于改进的动态用户行为模型得到的不同类型用户的状态转移模式存在显著差别，且具有更高的可解释性。

Abstract: Establishing user behavior models based on clickstream data and analyzing user characteristics are the basis for companies to develop accurate marketing strategies and provide personalized recommendations. Firstly, based on the traditional dynamic user behavior model (CBMG), this paper proposes an improved user behavior model that considers both page type sequences and behavior sequences to fully reflect its behavior patterns from user path preference information. Secondly, the user session is clustered based on the behavior sequence and the page type, and the session categories of different behavior patterns are obtained. Based on the Markov chain, the dynamic model of user behavior state transition is obtained among different conversation types. The research shows that the state transition patterns of different types of users based on the improved dynamic user behavior model are significantly different and have higher interpretability.