SP500215 Latent Semantic Indexing (LSI) and TREC-2 chapter S. Dumais National Institute of Standards and Technology D. K. Harman 5. Onward to TREC-3 We were quite pleased that we were able to use many of the existing LSIISVD tools on the ThEC-1 and TREC-2 collections. The most Important findmg in this regard was that the large, sparse SVD problems could be computed without numerical or convergence problems. We modified the preprocessing substantially for [OCRerr]fl[OCRerr]C-2, now bave many of the basic tools in place and should be able to conduct more experiments comparing various indexing and query matching ideas using the same underlying LSI engine. Bigger SVDs, faster query matching, Improving precision, and interactive interface issues are the major areas targeted for Improvement. [8] Dumais, S. T. and Schmitt, D. 0. Iterative searcbing in an oniine database. In Proceedings of Human Factors Society 35th Annual Meeting, 1991, 398402. [9] Evans, D., Leffe[OCRerr], R., Grefenstette, 0., Handerson, S., Hersh, W., and Archbold, A. CLM[OCRerr] fl[OCRerr]EC Design, experiments and results. lii D. Harman (Ed.) The First Text REtrieval Conference (TREC-1), NIST Special Publication 500-207, 1992,251-286. [10] Foltz, P. W. and Dumals, S. T. Personalized information delivery: An analysis of information filtering methods. Communications of the ACM, Dec. 1992, 35(12), 51-60. 6. References [1] Berry, M. W. computations. Supercomputer 49. Large scale singular value International Journal of Applications, 1992, 6(1), 13- [2] Buckley, C., Allan, J., and Salton, 0. Automatic routing and ad-hoc retrieval using SMART: IREC 2. To appear in: Proceedings of TREC-2. [3] Buckley, C. and Salton, 0. Automatic retrieval with locality information using SMART. In D. Harman (Ed.) The First Text REtrieval Conference (TREC-1), NIST Special Publication 500-207, 1992,59-72. [4] Cullum, JX. and Willoughby, R.A. Lanczos algorithms for large symmetric eigenvalue computations - Vol 1 Theory, (Chapter 5: Real rectangular matrices). Brii:hauser, Boston, 1985. [5] Deerwester, S., Dumais, S. T., Landauer, T. K., Furnas, 0. W. and Harshinan, R. A. Indexing by latent semantic analysis. Journal of the Society for Information Science, 1990, 41(6), 391-407. [6] Dumais, S. T. Improving the retrieval of information from exteiiial sources. Behavior Research Methods, Instruments and Computers, 1991,23(2), 229-236. [7] Dumais, S. T. LSI meets ThEC: A staus report. In D. Harman (Ed.) The First Text REtrieval Conference (TREC-1). NIST special publication 5O([OCRerr]207, 137-152. 114 [11] Furnas, 0. W., Deerwester, S., Dumals, S. T., Landaner, T. K., Harshman, R. A., Streeter, L. A., and Lochbaum, K. E. Information retrieval using a singular value decomposition model of latent semantic structure. In Proceedings of SIGIR, 1988,465-480. [12] Gallant, S., Hecht-Nielson, R., Cald, W., Qing, K., Carleton, J., Sudbeck, D. TIPSThR Panel - `INC's MatchPlus System. In D. Harman (Ed.) The First Text REtrieval Conference (TREC-1), NIST Special Publication 500-207, 1992, 107-112. [13] Jacobs, P, Krupka, 0., and Rau, L. A Boolean approxImation method for query construction and topic assignment in TREC. In D. Harman (Ed.) The First Text REtrieval Conference (TREC-1), NIST Special Publication 500-207, 1992,297-308. [14] Kane-Esrig, Y., Streeter, L., Dumais, S. T., Keese, W. and Casella, G. The relevance density method for multi-topic queries in information retrieval. In Proceedings of the 23rd Symposium on the Interface, E. Keramidas (Ed.), 1991,407-410. [15] Nelson, P. Site report for the Text REtrieval Conference. In D. Harman (Ed.) The First Text REtrieval Conference (TRFC-1), NIST Special Publication 500-207, 1992,287-2%. [16] Salton, 0. and McGill, M.J. Introduction to Modern Information Retrieval. McGraw-Hill, 1983. [17] Voorhees, E. On expanding query vectors with lexically related words. To appear in: Proceedings of TREC-2.