Using Hidden Markov Model to Predict the Web Users’ Linkage

摘要

Hidden Markov Model (HMM) has been proved to be an effective model in web usage mining. However, whether or not it works for all kinds of web site structure is still an important research issue. In this paper, we will analyze why web site should be divided into two types, that is, vertical structure and horizontal structure in detail. For different website structure we should adopt different approaches. While the traditional HMM approaches only works well for vertical structure website, however access prediction is more necessary for horizontal structure website. Herein we proposed a mining approach for horizontal website structure based on hidden Markov model. The approach is used to mine web user’ interest association rules and further predict the pages they may access. We conducted experiments using real web log data and demonstrated that our approach improved the prediction results of the horizontal structure website. The issue we investigated is useful in several areas, such as prefetching, personalization service, site modification, system improvement, business intelligence, etc.?