ART Neural Network based Clustering Agent for User Access Patterns

G.T. Raju and S. Khandige (India)


Knowledge Discovery, ART Neural Networks, Clustering, Web Usage mining


Knowledge about navigation patterns occurring in or dominating the usage of a website can greatly help the site’s owner or administrator in improving its quality. Data mining and AI techniques can assist in this task by effectively extracting knowledge from the past access recordings. Knowledge Discovery from the secondary data generated by the user interactions with the web has become very critical for effective and efficient managing of the activities related to e-business, e-services-education, personalization, web site management and so on. Mining Web access logs that contain substantial data about user access patterns on one or more web localities is an emerging research area. One of the important phases in mining user access patterns is the clustering of web users. In this paper, we present an approach to dynamically group web users based on their web access patterns using Adaptive Resonance Theory Neural Network. Knowledge extracted from web user clusters has been used for pre fetching of pages between web clients and proxies. Experiments have been conducted and the results have shown that our ART based clustering approach performed better in terms of intra-cluster distances. The degree of personalization that a web site offers in presenting its services to users is an important attribute contributing to the site’s popularity. Web usage mining works on user profiles, user access patterns and navigation paths that are being heavily used by e commerce companies, for tracking customer behavior on their sites. Restructuring of sites to individual user interests increases the computation at the server to an impractical degree. One way of solving this problem is to group users based on their web interests and then organize the structure of site according to the needs of different groups. Personalized Web agents include those that obtain or learn user preferences and discover web information sources that correspond to these preferences and possibly those of other individuals with similar interests (using collaborative filtering). Ex: Web watcher, which utilizes the user profile and learn to rate web pages of interest using Bayesian classifier. We have developed an algorithm to cluster users according to their access patterns based on ART neural network that offers an unsupervised clustering. This approach adapts to changes in user access patterns over time without losing earlier information. Each cluster is represented as prototype vector by generalizing the URLs most frequently accessed by all cluster members.

Important Links:

Go Back