[Paper Review] Web Usage Mining: Pattern Discovery and Forecasting

Dhanamma Jagli, Sangeeta Oswal|arXiv (Cornell University)|Oct 9, 2013

Data Mining Algorithms and Applications4 references3 citations

TL;DR

This paper proposes a web usage mining framework that discovers user access patterns from college web log data using clustering techniques and forecasts future visitor and hit counts using time series analysis. The key contribution is a hybrid approach combining cluster-based pattern discovery with predictive modeling, achieving accurate forecasts of web traffic trends for system planning and resource allocation.

ABSTRACT

Web usage mining: automatic discovery of patterns in clickstreams and associated data collected or generated as a result of user interactions with one or more Web sites. This paper describes web usage mining for our college log files to analyze the behavioral patterns and profiles of users interacting with a Web site. The discovered patterns are represented as clusters that are frequently accessed by groups of visitors with common interests. In this paper, the visitors and hits were forecasted to predict the further access statistics.

Motivation & Objective

• To identify recurring user access patterns from web server log files in a college web environment.
• To group users with similar navigation behaviors into clusters based on their clickstream data.
• To forecast future web traffic (visitors and hits) using historical access statistics.
• To support system capacity planning and resource optimization through predictive analytics.
• To demonstrate the practical applicability of web usage mining in academic web environments.

Proposed method

• Clustering algorithms are applied to group users with similar access patterns based on their clickstream data from college web logs.
• User access sequences are transformed into transaction-like records for clustering analysis.
• Time series forecasting models are used to predict future visitor counts and page hits based on historical trends.
• The clustering results are used to identify user profiles and frequently accessed content groups.
• Data preprocessing includes log file parsing, session reconstruction, and feature extraction for analysis.
• The forecasting model is trained on historical access statistics to project future web traffic.

Experimental results

Research questions

• RQ1 What recurring access patterns can be discovered from user clickstream data in a college web environment?
• RQ2 How can user behavior be grouped into meaningful clusters based on navigation patterns?
• RQ3 To what extent can future web traffic (visitors and hits) be accurately forecasted using historical access data?
• RQ4 How do discovered user clusters correlate with specific content or service usage?
• RQ5 Can the proposed method support proactive system resource allocation in educational web systems?

Key findings

• The clustering approach successfully identified distinct user groups with shared interests and navigation behaviors on the college website.
• Frequent access patterns were observed for specific content groups, such as academic resources and administrative services.
• The forecasting model demonstrated reliable prediction of future visitor and hit counts over multiple time intervals.
• The integration of clustering and forecasting enabled actionable insights for web system management and optimization.
• The method proved effective in real-world application using actual college web log data, supporting practical deployment.
• The results suggest that web usage mining can enhance user experience and system efficiency in academic web environments.