Personalisierte Volltextsuche mit partiellen Ordnungen

Abstract

Recently, the importance of personalization as a tool against information overload, experienced for example by users surfing the internet, increases. The intention behind personalization is the individual filtering and preparation of information. In this context, very good experiences were made at the "Lehrstuhl für Datenbanken und Informationssysteme" of the University of Augsburg with attribute-based queries formulated as partial orders, which are called "preferences". Consequently, this work investigates partial orders for full-text searches. Two alternatives for constructing such preferences exist within the "Preference World" framework: base preferences, which define a partial order on the documents, and preference constructors, which combine preferences. With the so called "subspace preference", the approach using preference constructors proved promising. The subspace preference divides a query into several subqueries, which are evaluated as vector space models. The results are then combined by cumulation, which treats the subqueries as equally important, or prioritization, which preferes a subquery over another. This way, users have the opportunity to provide additional semantics easily -- without beeing forced to do so, since the vector space model is still contained as a special case. An evaluation of the subspace preference indicates that the subspace preference is capable of transforming these additional semantics into better retrieval results.