Efficient Mining of Web Access Patterns using Constrained Self-Organizing Map Clustering

Abstract

Web usage mining attempts to reveal interesting patterns of web access from a large number of web users. The main source of web usage data is web server logs. From an organizational point of view, web usage mining is very important as it assists in server management. The administrators of web servers can analyze web server log data to understand user behavior, allocate resources accordingly and provide customized service to similar groups of users. Clustering is a predominant data mining that partitions a group of unlabelled data instances into distinct groups or clusters. Several clustering techniques have been proposed in literature, which includes stand-alone as well as ensemble clustering techniques. Most of them lack robustness and cannot effectively visualize clustering results to help knowledge discovery and constructive learning. This paper explains the use of Self Organizing Maps (SOM) in a cluster ensemble framework based on some prior input constraints. Cluster ensemble is a set of clustering solutions obtained as a result of individual clustering on subsets of the original high-dimensional data. The final consensus matrix is fed to a neural network which transforms the input data to a lower-dimensional output map. The map clearly depicts the distribution of input data instances into clusters. The proposed method is tested on real web log data. Evaluation of clusters obtained justifies the superiority of the approach over conventional clustering techniques.