WebPersonalizer: A Server-Side Recommender System Based on Web Usage Mining

Existing approaches to Web personalization often rely heavily on explicit and subjective user input resulting in static profiles which are prone to biases. In this paper we present a usagebased Web personalization system, called WebPersonalizer, drawing heavily upon Web mining techniques, making the personalization process automatic, and dynamic. The system architecture separates the offline tasks of data preparation and Web usage mining, and the online recommendation engine. At the heart of the system is a technique based on clustering of user transactions which allows for the discovery of effective aggregate usage profiles. We discuss how the discovered aggregate profiles can be used in conjunction with the current status of an ongoing Web activity to perform real-time personalization. The Web usage mining approach allows a site to provide effective personalization using anonymous and implicit user behavioral patterns without relying on subjective or personally identifying user input.