Web Usage Mining and Online Recommendations

Web Usage Mining analyses the usage patterns of web sites in order to get an improved understanding of the users’ interests and requirements. This information is especially valuabale for E-Business sites in order to achieve improved customer satisfaction. Our approach is based on a Web Usage Data Warehouse to provide scalability and allow for flexible multi-dimensional analysis and data mining. Furthermore, it can be coupled with business warehouses, e.g. for customer relationship management. We developed a prototype implementation for our own web sites using commercially available tools.

We currently focus on the application of web usage mining for automatically determining Web Recommendations. Recommendations help users to quickly find the information they want or find interesting. On the other hand, they allow website owners to optimize the website, increase web user satisfaction and save on the costs of content management. An overview of our recommendation architecture is shown below. Recommendations are dynamically determined either based on manually specified rules or automatically determined by different recommendation algorithms. Our approach evaluates the effectiveness of shown recommendations using Machine Learning algorithms and thus exploits user feedback to dynamically select the optimal recommendations. Data Warehouse technology is used to effectively manage large amounts of usage data and support various recommender algorithms

We recently developed the AWESOME (Adaptive Website Recommendations) prototype, which captures and evaluates the recommendation feedback. The data warehouse technology enables OLAP reports for detailed performance studies of all recommenders regarding different contexts like the current page or referrer information. Moreover, AWESOME performs an automatic and adaptive closed-loop website optimization by dynamically selecting the most promising recommenders based on continuously measured recommendation feedback. First results show that good recommender selection strategies increase the acceptance rate of presented recommendations. We are always interested to apply AWESOME to other websites to verify our results.

Project Members:

Selected Publications

Thor, A.; Golovin, N.; Rahm, E. Adaptive Website Recommendations with AWESOME VLDB Journal 14(4), 2005 2005-11

Golovin, N.; Rahm, E. Automatic Optimization of Web Recommendations Using Feedback and Ontology Graphs Proc. 5th Int. Conf. on Web Engineering (ICWE), Springer LNCS 3579, 2005: 375-386 2005-09

Golovin, N.; Rahm, E. Reinforcement Learning Architecture for Web Recommendations Proceedings of the ITCC 2004 2004-12

Thor, A.; Rahm, E. AWESOME - A Data Warehouse-based System for Adaptive Website Recommendations Proc. 30th Intl. Conference on Very Large Databases (VLDB), 2004 2004-08

Thor, A.; Rahm, E. Data-Warehouse-basierte Architektur für adaptive Online-Recommendations Proc. Workshop “Content- und Wissensmanagement”, Leipziger Informatiktage (LIT), 2003 2003-09

Rahm, E.; Vossen, G. (Hrsg.) Web & Datenbanken 2002-09

Rahm, E. Web Usage Mining. Datenbank-Spektrum Vol. 2, Heft 2, 75-76 (Feb. 2002) 2002

Rahm, E., Stöhr, T. Data Warehouse-Einsatz zur Web-Zugriffsanalyse. In: Web & Datenbanken. Dpunkt-Verlag, 2003 2002

Stöhr, T.; Rahm, E.; Quitzsch, S. OLAP-Auswertung von Web-Zugriffen Proc. GI-Workshop Internet-Datenbanken, Berlin, Sep. 2000 2000

Master theses:

Kötz, M.: Werkzeuge zur statistischen Auswertung von Web-Zugriffen Jul 2002

Quitzsch, S.: Metadaten-Nutzung kommerzieller Data Warehouse-Werkzeuge, April 2000