Preference Mining and Preference Repositories: Design, Algorithms and Personalized Applications

Stefan Holland

Preference Mining and Preference Repositories: Design, Algorithms and Personalized Applications

Dissertation, University of Augsburg.
1st Examiner: Professor Dr. W. Kießling
2nd Examiner: Professor Dr. M. Ester
erschienen 12/2003


Advanced personalized e-applications require comprehensive knowledge about their user's likes and dislikes in order to provide individual product recommendations, personal customer advice, and custom-tailored product offers. Modeling such preferences as strict partial orders with "A is better than B" semantics has proven to be very suitable in various e-applications.
In this thesis we developed novel Preference Mining techniques for detecting strict partial order preferences in user log data. Categorical and numerical preferences are discovered based on the frequencies of the different attribute values in the user log. These base preferences are then combined to detect complex preferences. The main advantage of this approach is the semantically rich and easily interpretable Preference Mining result. Extensive experimental evaluations on real and simulated data proved the effectiveness and efficiency of our algorithms.

In this work we also developed a Preference Repository, which is an XML-based storage structure for preferences. Thereby, it is not important whether the preferences are detected with Preference Mining or have any other origin like preference queries or preference questionnaires. The Preference Repository stores not only the preferences themselves but also relevant meta information like timestamp, user identifier, or underlying situation.
The integration of situations is essential because user preferences frequently do not hold in general but may depend on underlying situations. Therefore, we introduced the concepts of situation modeling and showed how situations can be integrated into entity-relationship models. Several case studies proved the usability of this approach for personalized applications that have to care for their users' situations.

Our Preference Repository also provides a set of access operations for reading, writing, and updating preferences as well as advanced operations that are useful for personalized applications. These include not only the computation of preferences that match best to given situations but also the detection of user groups containing users that have the same preferences. The latter can be used to offer product recommendations based on the preferred products of other users in the same group.

With our innovations on Preference Mining and Preference Repositories personalized e-applications can gain valuable knowledge about their customers' preferences, which is essential for a qualified customer service.