Personalized Keyword Search with Partial-Order Preferences

Personalized search engines must be able to cope with various user preferences, retrieving the best matches to a query. For SQL and XML applications new methods for such preference-based searches have been implemented recently. Here we adopt this approach to keyword search in full-text search engines. We propose to augment the vector space model (VSM) by preference constructors having an intuitive partial order semantics: Pareto-accumulation and prioritization. We show that prioritization can be interpreted as subspace preference in the VSM. Using a preprocessor approach we succeed to map prioritization onto the VSM. A first query benchmark, using the standard Time-collection, revealed promising results. The retrieval quality, measured by average expected search length, could be slightly improved. Using the proposed meta engine approach, this gain in retrieval quality is accompanied by a substantial speed up of query runtimes. Thus subspace preferences can be integrated efficiently into existing full-text search engines.