Adjustment of indirect association rules for the web

Abstract

Indirect association rules are the extension of classic association rules that enables to discover indirect relationships existing between objects. To estimate the importance of individual parameters of the indirect association rules mining, experiments were carried out on historical web user sessions coming from an e-commerce portal. The influence of parameters of standard direct rules: direct support and direct confidence thresholds, was studied and it was proved that greater values of these two thresholds could significantly decrease the final quantity of indirect rules. This reduction may be additionally strengthened by the introduction of additional threshold to complete or partial indirect confidence. The choice of calculation method for partial indirect confidence was also examined and the multiplication method was selected as the most discriminative. © Springer-Verlag Berlin Heidelberg 2004.