Web usage mining: knowledge discovery using Markov chains

The main focus of the work is presenting a modeling technique to develop a dynamic and interactive-tool that is able to improve the knowledge of how a web site is being used and consequently to support the activity of site designing and restructuring. The present work proposes a new modeling technique based on Markovian assumption to be applied to data contained in the log file. A framework is defined whose states are described by vectors and the transitions among such states are modeled through Markov chains. The results of the modeling step are accessed through a graphical user interface that gives a concise and dynamic representation of the navigation paths in a web site and can be used interactively to carry on several kinds of analysis. The paper presents some examples of information that can be extracted through this method and suggests some possible application in decision supporting. The paper suggests a further improvement to the Markovian analysis based on dynamic profiling of the site users by means of a clustering activity.