P Roceedings of the Ecml / Pkdd ’ 2007 D Iscovery C Hallenge

This task is co-organized by Gemius SA, the leading Internet market research agency in Central and Eastern Europe. The problem objective is to predict user behaviour by characterising nature of user’s visit, i.e., the list categories of the visited Internet portal and the number of page views in each category. The challenge is accomplished with use of web trafic data from Polish web sites employing gemiusTraffic study, grouped by appropriate categories. This will be accomplished with use of web traffic data from Polish web sites employing gemiusTraffic study, grouped by appropriate categories. The above defined objective has been divided into three separate challenge problems: Problem 1 is related to the length of the visit. A visit – accordingly to the definition – is a sequence of page views by one user (cookie). As web pages are identified by their categories, during one visit user may view pages of one or more categories. The problem is to predict whether a given visit is short (1 category) or long (two or more categories). Problem 2 is related to the most probable categories. Solution of Problem 2 is a list of the most probable categories in a given visit of a given user. Problem 3 is to predict the most probable categories and ranges of numbers of page views. Solution of Problem 3 is a list of the most probable categories in a given visit of a given user with range of number of page views in each category.