Concept-based IR is expected to improve the quality of medical ranking since it captures more semantics than BOW representations. However, bringing concepts and BOW together into a transparent IR framework is challenging. We propose a new aggregation parameter to combine conceptual and term-based Dirichlet Compound Model scores effectively. The determination of this linear parameter is the result of exploring to what degree the difference of the conceptual and term-based sum of IDFs is influential to the integration. Instead of employing heuristics to find combined models, this paper aims to build the grounds for establishing reasonable aggregation standards based on semantic query performance predictors.
Authors: Mohammad Bahrani, Thomas Roelleke