Characterizing Visitors to a Website Across Multiple Sessions

Characterizing web users based on their interactions with a particular website is a key problem in web mining. Such interactions are reflected by clickstream data in weblogs, the content of the pages that they view, as well as their actions such as search queries typed in. In general a wide variety of behaviors is observed at popular websites. So, to make the problem more tractable, it is often wise to first group similar users into relatively homogeneous segments and then try to characterize each such segment. There has been much work done on clustering visitors at a site, but typically the analyses use substantially simplified descriptions of session behavior, and also do not characterize users based on their footprints across multiple visits to the site. In this paper, each web user is represented by a set of sessions, and a two stage methodology for grouping such users is proposed. In the first stage, sessions are clustered into conceptual session types based on both the trajectory taken through the website as well as the time spent at each page. The distribution of a user’s multiple visits across these session types forms the basis of user grouping in the second stage. Such clustering based on a rich user description forms a platform for further characterization. Results are presented using weblogs of a popular website to illustrate the techniques.