Entry Vocabulary - a Technology to Enhance Digital Search

Entry Vocabulary – a Technology to Enhance Digital Search Fredric Gey, Michael Buckland, Aitao Chen and Ray Larson University of California Berkeley, CA 94720 gey@ucdata.berkeley.edu, faitao,buc kland,ra yg@sims.b erk eley .edu ABSTRACT This pap er describ es a searc h tec hnology whic h enables im- pro v ed searc h across div erse genres of digital ob jects { do c- umen ts, paten ts, cross-language retriev al, n umeric data and images. The tec hnology lev erages h uman indexing of ob- jects in sp ecialized domains to pro vide increased accessibil- it y to non-exp ert searc hers. Our approac h is the rev erse- engineer text categorization to supply mappings from ordi- nary language v o cabulary to sp ecialist v o cabulary b y con- structing maxim um lik eliho o d mappings b et w een w ords and phrases and classi�cation sc hemes. This forms the train- ing data or 'en try v o cabulary'; subsequen tly user queries are matc hed against the en try v o cabulary to expand the searc h univ erse. The tec hnology has b een applied to searc h of paten t databases, n umeric economic statistics, and for- eign language do cumen t collections. 1. INTRODUCTION The in ternet has pro vided a v ast and gro wing amoun t of searc hable information. In the "deep w eb" (that part of the in ternet whic h is not directly searc hable using ordinary searc h engines) w e �nd information deriving from m ultiple and quite distinct genres, suc h as images and n umeric sta- tistical data as w ell as textual information in unfamiliar lan- guages. F or example United States F oreign T rade Imp orts and Exp orts are a v ailable at http://govinfo.kerr.orst.e du/imp exp.html. Data are classi�ed b y commo dit y b eing shipp ed, so one can �nd out, for example, ho w man y purebred Arabian horses w ere imp orted to Louisville, Ken tuc ky from Saudi Arabia in an y particular mon th. A commo dit y searc h mec hanism is pro vided at this sites to searc h commo dit y descriptions asso ciated with the 8,000 commo dit y co des. Ho w ev er, the searc h term 'automobile' retriev es nothing, ev en though w e kno w billions of U.S. dollars of automobile imp orts en ter the United States eac h y ear. In order to retriev e automobile im- p orts with using the string searc h one needs to kno w that the description is actually "P ass Mtr V eh" an abbreviation . for "P assenger Motor V ehicle" and obtain the data sho wn in Figure 1. In another case, supp ose a searc her from German y is in- terested in articles on economic p olicy and wishes to searc h in his/her nativ e language. The searc h term "Wirtsc …

Content truncated to 500 words