Web-user Identification through Webpage Session Tracking

This research paper has gone some way towards enhancing the understanding of the application of Web Usage Mining methods for a user identification based upon a user's profile in web browsing. The aim of this study is to propose a model using Data Mining techniques for a user identification based on web browsing history. The specific objectives of the paper are to: 1. Study the theoretical aspects of Web Usage Mining. 2. Determine the suitable options for data modeling and feature extraction. 3. Choose the clustering algorithm for web usage pattern discovery. 4. Define the optimal metrics and success criteria for model assessment. 5. Perform a comparative analysis of multiple classifiers. The data in this work consists of processed web server log records of 3370 users interacting with web resources in a 6-month period. This research paper outlines a novel method of a user identification based on the conjunction of traditional Data Mining, Web Mining and Text Mining techniques. The architecture of the model consists of three main stages defined by CRISP-DM methodology: data understanding, data preparation (which technically represents preprocessing and feature engineering) and modeling. The devised methodology not only focuses on different approaches to data modeling and ensemble learning, specifically stacking of several linear models with gradient boosting metamodel, but also incorporates the latest technological advances of online-learning for handling large data volume. In addition, during the cluster analysis there have been retrieved the aggregated profiles exhibiting the most common web browsing patterns. The present findings have important implications for solving the problem of user identity and could be a stable baseline for a more complex and sophisticated approach, that might be used for monitoring account security as well as for recommendation or personalization services improvement. Key words: Web Usage Mining, machine learning.