Lazy Query Expansion

An information retrieval or document base system has to somehow deal with various phenomena of equivalence of some strings. These are lowercase versus uppercase matching, morphological inflection, derivation, and synonymy of words: e.g., given a query computer, find Computers, computing, workstation. The latter problems are very important in languages with richer morphology and less stable terminology than in English. Also, much better recall is achieved by matching hyponyms and hypernyms using a thesaurus, e.g., given a query computers, find also supercomputer, microcomputer, mainframe, machine, device, processor, UNIX, etc. Technically, this can be handled at the time of indexing by reducing related strings to a common form, or at the time of query processing by expanding the query with the whole set of the related forms. We argue for that the latter way allows for greater flexibility and easier maintenance, while being more affordable than it is usually considered. We propose to expand the query with only those words that really appear in the document base. Our experiments with a thesaurus-based information retrieval system we are developing for the Senate of Mexican Republic show only insignificant increase of the real user queries on average with the 200-megabyte document base of the Senate, in spite of highly inflective Spanish language.