Variable Length Markov Chains for Web Usage Mining

View Sample PDF

Author(s): José Borges (School of Engineering, University of Porto, Portugal)and Mark Levene (Birkbeck, University of London, UK)
Copyright: 2009
Pages: 5
Source title: Encyclopedia of Data Warehousing and Mining, Second Edition
Source Author(s)/Editor(s): John Wang (Montclair State University, USA)
DOI: 10.4018/978-1-60566-010-3.ch310

Purchase

View Variable Length Markov Chains for Web Usage Mining on the publisher's website for pricing and purchasing information.

Abstract

Web usage mining is usually defined as the discipline that concentrates on developing techniques that model and study users’ Web navigation behavior by means of analyzing data obtained from user interactions with Web resources; see (Mobasher, 2006; Liu, 2007) for recent reviews on web usage mining. When users access Web resources they leave a trace behind that is stored in log files, such traces are called clickstream records. Clickstream records can be preprocessed into time-ordered sessions of sequential clicks (Spiliopoulou et al., 2003), where a user session represents a trail the user followed through the Web space. The process of session reconstruction is called sessionizing. Understanding user Web navigation behavior is a fundamental step in providing guidelines on how to improve users’ Web experience. In this context, a model able to represent usage data can be used to induce frequent navigation patterns, to predict future user navigation intentions, and to provide a platform for adapting Web pages according to user specific information needs (Anand et al., 2005; Eirinaki et al., 2007). Techniques using association rules (Herlocker et al., 2004) or clustering methods (Mobasher et al., 2002) have been used in this context. Given a set of transactions clustering techniques can be used, for example, to find user segments, and association rule techniques can be used, for example, to find important relationships among pages based on the users navigational patterns. These methods have the limitation that the ordering of page views is not taken into consideration in the modeling of user sessions (Liu, 2007). Two methods that take into account the page view ordering are: tree based methods (Chen et al., 2003) used for prefetching Web resources, and Markov models (Borges et al., 2000; Deshpande et al., 2004) used for link prediction. Moreover, recent studies have been conducted on the use of visualization techniques for discovering navigational trends from usage data (Chen et al., 2007a; Chen et al., 2007b).

The IRMA Community

Research IRM

Variable Length Markov Chains for Web Usage Mining

Purchase

Abstract

Related Content

IRMA Sponsors