Incorporating Concept Hierarchies into Usage Mining

Abstract

Web usage mining is being used extensively for Web Personalization. Many algorithms and techniques have been proposed to predict the next user request. Most, however, are limited in terms of their ability to use concept hierarchy and connectivity of the website. Recent studies have shown that conceptual and structural characteristics of the website play an important role in the quality of the recommendation models. In this paper we propose a new technique to incorporate conceptual characteristics of a website into the recommendation models, and use sequence alignment techniques, adapted from the field of bioinformatics, coupled with a new model for defining page similarity. We introduce a scoring methodology to quantify page similarity derived from the concept hierarchy of a website. These scores are an essential ingredient in the sequence alignment technique. Other aspects, like time spent by the user on a page and page access sequence are also considered during the alignment. Thus, the system that we propose makes use of various sources of information to make recommendations. Finally we present experimental results to show the effectiveness of our method.