This research will develop a methodology for utilizing a lexical database in a statistical retrieval model to improve retrieval performance (i.e., recall and precision). The sesame architecture proposed in this proposal will utilize the lexical database in four major phases of retrieval: (1) document analysis; (2) query formulation; (3) retrieval and ranking; and (4) query reformulation. A prototype will be built to verify the effectiveness of sesame using a set of document databases developed at the Cornell University. Wordnet, a large lexical database developed at Princeton University, will be used in this project. The anticipated benefit is a methodology which will improve the effectiveness of methods based on purely Boolean keyword search or the statistical model. In the phase I research, the methodology will be developed and a prototype will be built. The results from this research will have huge market potential in personal information management systems, office information systems, library systems, and wide-area information servers.
