Using External Knowledge to Solve Multi-Dimensional Queries

To improve the precision of an information retrieval system in a specific domain we propose a new indexing scheme based on external knowledge resources such as thesauri or ontologies. We introduce the notion of domain dimension, which is a substructure of a knowledge resource, to formally represent the different aspects of a domain that appear in a document. Then, we identify dimensions in documents and queries using a conceptual indexing. The result of this indexing is a representation of each document along its semantic dimensions. We also propose a query processing based on multi-dimensional indexing. It is comprised of a dimensional filtering followed by a dimensional ranking. Experimental results on medical imaging documents (ImageCLEFmed-2005 collection) show that the dimensional filtering, using three dimensions, can improves the mean average precision by about 25%.