Incremental Cluster Interpretation with Fuzzy ART in Web Analytics

Abstract

Clustering in web analytics extracts information from data based on similarity measurement on the data patterns, where similar data patterns are grouped as a cluster. However, the typical clustering methods used in web analytics suffer from three major shortcomings, viz., (1) the predefined number of clusters is hard to determined when new data are generated over time; (2) new data might not be adopted into the existing clusters; and (3) the information given by a cluster (centroid) is vague. In this study, an incremental learning method using the Fuzzy Adaptive Resonance Theory (Fuzzy ART) algorithm is adopted (1) to analyze the underlying structure (hidden message) of the data, and (2) to interpret cluster into an understandable and useful knowledge about user activity on a webpage. An experimental case study was conducted by capturing the integrated data from Google Analytics on the University of Technology Sarawak (UTS), Malaysia, website to analyze user activity on the webpage. The results were analyzed and discussed, and it shown that the information obtained at each cluster can be interpreted in term of cluster boundary at each feature space (dimension), whereas the user activity are explained from the cluster boundary without revisiting the trained data.