Analysis of Web Visit Histories, Part II: Predicting Navigation by Nested STUMP Regression Trees

This paper constitutes part II of the contribution to the analysis of web visit histories through a new methodological framework for web usage-structure mining considering association rules theory. The aim is to explore through a tree structure the sequence of direct rules (i.e. paths) that characterize a web navigator who keeps standing longer on a web page with respect to the path characterizing navigators who leave the web earlier. A novel tree-based structure is introduced to take into account that the learning sample changes click by click leaving out navigators who drop off from the web after any click. The response variable at each time point is the remaining number of clicks before leaving the web. The split is induced by the predictors that describe the preferred web sections. The methodology introduced results in a Nested Stump Regression Tree that is an hierarchy of stump trees, where a stump is a tree with only one split or, equivalently, with only two terminal nodes. Suitable properties are outlined. As in first part of the contribution to the analysis of the web visit histories, a methodological description is provided by considering a web portal with a fixed set of web sections, i.e. a data set coming from the UCI Machine Learning Repository.