SIGIR 2005 Doctoral Consortium
David J. Harper · 2005
We address the problem of developing a flexible framework for information retrieval (IR) in structured documents, such as XML. The framework is able to support a wide range of structured IR queries, transparent instantiations of different retrieval models, and different physical implementations. It is based on so-called score region algebra (SRA) that can express the following four essential ranked retrieval aspects for structured IR: term and element selection, element relevance score computation, element score propagation, and element score combination. Our preliminary research shows that different instantiations of each aspect, as well as different combinations of these instantiations, yield significantly different results. Our goal is to better understand structured IR by studying these aspects alone and their combination in the framework of SRA, and to use this knowledge to improve our structured IR system.