Multiple Queries of Information Retrieval Using Krylov Subspace Method

The Krylov subspace based information retrieval (IR) approach has been shown to provide comparable accuracy to latent semantic indexing (LSI), while providing some computational advantages. Recently, in the area of numerical linear algebra, attention has been drawn to the block Krylov subspace methods, which are shown to be more efficient than the classic Krylov subspace methods in solving linear systems with multiple right hand sides. Such improvement in the algorithm gives us the opportunity to extend the original retrieval method, enabling single query searching, to multiple query searching. In this paper, we report such improvement in the retrieval algorithm, and demonstrate its performance by comparing to several other retrieval methods using the Medline corpus.