Clustering of users on web log data using Optimized CURE Clustering

Author Name(s): *P. Dhanalakshmi, K. Ramani, B. Eswara reddy
Author Email:

Abstract

Web usage mining is one of the essential frameworks to find domain knowledge from interaction of users with the web. This domain knowledge is used for effective management of predictive websites, creation of adaptive websites, enhancing business and web services, personalization and so on. In non-profitable organization’s website it is difficult to identify who are users, what information they need, and their interests change with time. Web usage mining based on log data provides a solution to this problem. The proposed work focuses on clustering of web users on web log data by developing an optimized clustering algorithm. The performance of web usage mining is also compared based on k-means, DBSCAN and CURE with proposed algorithm.

Keywords

Clustering, K-Means, Density based clustering, Optimized clusters, Heap tree, KD tree, CURE Clustering, Manhattan distance measure, weblog, web usage mining.

Introduction

Data Mining helps to extract only relevant information from these large repositories. Web is a huge repository of text documents and multimedia data. Mining useful data from the web is known as web mining and it may include analysis of web content, structure and also understanding usage of web . Usage of internet has magnified number of web users enormously increasing web log data.  To understand the web user behavior one has to study web log data. It has three stages such as preprocessing, identifying usage patterns and analyzing the obtained patterns. Web server accounting all users’ actions of the web site as web servers Logs. The majority of  the log files have Common log file (CLF) format and each log contains IP Address, user name, date, time stamp, visited URL, Request type. web log mining is a process of  analyzing and discovering user access patterns and their behavior using  weblog data which is available may be in proxy, web servers or client browsers. Clustering of these web users in E-commerce sites becomes an essential approach for many applications: web site personalization and structure optimization, advertising purposes and reconstruction of websites.  The web log data obtained  by a web server can be applied to clustering process, where the users are

Conclusion

In this paper, an improved CURE Clustering approach using Normal Mixture distribution model was implemented to cluster the web users on weblog dataset. Normal mixture model is used to transform the web log data as initial clusters. In this approach, KD tree followed by heap tree data structures are used to represent the clusters and representative points. In this model, we have optimized the initial clusters representation, merging process and distance measure of traditional CURE clustering algorithm. Also, Manhattan distance measure is used to find the nearest distance between the heap objects with merge operation to determine closest clusters. Experimental results show that proposed model has high computation efficiency in terms of cluster quality precision and error rate compared to traditional models. In future, this work can be extended to improve the real-time web log processing using parallel programming.

1,500 total views, no views today

Download PDF File

About the author: tej