Applying the User-over-Ranking Hypothesis to Query Formulation

The User-over-Ranking hypothesis states that the best retrieval performance can be achieved with queries returning about as many results as can be considered at user site [21]. We apply this hypothesis to Lee et al.’s problem of automatically formulating a single promising query from a given set of keywords [16]. Lee et al.’s original approach requires unrestricted access to the retrieval system’s index and manual parameter tuning for each keyword set. Their approach is not applicable on larger scale, not to mention web search scenarios. By applying the User-over-Ranking hypothesis we overcome this restriction and present a fully automatic user-site heuristic for web query formulation from given keywords. Substantial performance gains of up to 60% runtime improvement over previous approaches for similar problems underpin the value of our approach.