Collection-Relative Representations: A Unifying View to Retrieval Models

Various retrieval models have been developed and analyzed so far, but less research aims to an integration of the different models within a common framework. This paper introduces the idea of collection-relative retrieval models, a paradigm where several important retrieval models fit in. Our unifying view helps to better understand retrieval models, and it can be considered as a step towards a common theoretical framework for text retrieval. Collection-relative retrieval models employ a so-called index collection. We present an evaluation that shows how particular characteristics of the underlying index collection affect the retrieval performance of a collection-relative model. Based on such insights tailored index collections can be constructed in order to address specialized retrieval tasks.