Motivation: Acronyms result from a highly productive type of term variation and trigger the need for an acronym dictionary to establish associations between acronyms and their expanded forms. Results: We propose a novel method for recognizing acronym defin

Bioinformatics Advance Access published October 18, 2006

BIOINFORMATICS

BuildinganAbbreviationDictionaryUsingaTermRecognitionApproach

NaoakiOkazakia,b,SophiaAnaniadouc,d

a

GraduateSchoolofInformationScienceandTechnology,TheUniversityofTokyo,7-3-1Hongo,Bunkyo-ku,Tokyo113-8651,Japan.b

JapanSocietyforthePromotionofScience(JSPS).c

SchoolofComputerScience,TheUniversityofManchester,d

NationalCentreforTextMining(NaCTeM),

ManchesterInterdisciplinaryBiocentre,OxfordRoad,Manchester,M139PL.

Associate Editor: Golan Yona

ABSTRACT

Motivation:Acronymsresultfromahighlyproductivetypeoftermvariationandtriggertheneedforanacronymdictionarytoestablishassociationsbetweenacronymsandtheirexpandedforms.

Results:Weproposeanovelmethodforrecognizingacronymde nitionsinatextcollection.Assumingawordsequenceco-occurringfrequentlywithaparentheticalexpressiontobeapotentialexpandedform,ourmethodidenti esacronymde nitionsinasimilarmannertothestatisticalterm-recognitiontask.AppliedtothewholeMEDLINE(7,811,582abstracts),theimplementedsystemextracted886,755acronymcandidatesandrecognized300,954expandedformsinreasonabletime.Ourmethodoutperformedbase-linesystems,achieving99%precisionand82–95%recallonourevaluationcorpusthatroughlyemulatesthewholeMEDLINE.

AvailabilityandSupplementaryInformation:Theimplementationsandsupplementaryinformationareavailableatourwebsite:www.1mpi.com

1INTRODUCTION

Acronymsresultfromahighlyproductivetypeoftermvariationwhichsubstitutesfullyexpandedterms(e.g.,retinoicacidreceptoralpha)withshortenedterm-forms(e.g.,RARA).Changetal.(2006)reportedthat64,242newacronymswereintroducedin2004inMEDLINEabstracts.Terminologicalresourcesandscienti cdatabases(suchasUMLS1,Swiss-Prot2,SGD3,FlyBase4,andUniProt5)cannotkeepup-to-datewiththegrowthofneologisms(Pustejovskyetal.,2001).Inpractice,nogenericrulesorexactpatternshavebeenestablishedfordealingwithacronymcreation.Gaudanetal.(2005)distinguishedglobalacronymsfromlocalacronymsbasedonthepresenceoftheirde nitionsintexts.Globalacronymsappearindocumentswithouttheexpandedformexplicitlystated,whilelocalacronymsaccompanytheirexpandedformsinthedocument.Globalacronymshindertext-miningtaskssuchasinformationretrievalandinformationextraction.Wrenetal.(2005)

12345

reportedthatPubMedcouldretrieve5,477documentsforJNKbutonly3,773documentsforitsfullterm,c-junN-terminalkinase.Thus,anacronymdictionaryisnecessaryforadvancedtext-miningtaskstoestablishassociationsbetweenacronymsandtheirexpandedforms.Adar(2004)notedthatpreviousworkhadmostlyfoundacronymde nitionswithinatextinasimilarmannertoinformationextraction.Hesawtheneedforadditionaltasksforapracticalacronymresourcesuchasmergingsimilarde nitionsandprovidingdisambiguationinformation.Althoughwe ndsuchcomponentsindispensablefortext-miningapplications,herewefocusonthetaskof ndingacronymde nitions,asthe rststepinbuildinganaccurateacronymdictionary.

Anotherimportantaspectforbuildinganacronymdictionaryisthedistinctionbetweendynamicandcommonacronyms(Yuetal.,2002).Dynamicacronymsareone-timesubstitutionsvalidwithinadocumentandthereforealwayslocal.Incontrast,commonacronymsareusedovertwoormorepublications,andmayappearindocumentswithorwithouttheirexpandedforms.Anacronymdictionaryshouldfocusoncommonacronymssincetheyarepotentialglobalacronyms,i.e.,mightbewrittenwithouttheirde nitionsinsomedocuments.Inthispaperwedonotdealwiththeidenti cationofdynamicacronymswhichcanberecognizedbylettermatchingtechniques.Wecollectde nitionsoflocalandcommonacronymsinsourcedocuments.

Oneofthemainchallengesoftextminingisdealingwithanenormousamountofdocumentsinascalableandef cientmanner.Atthesametime,wecanalsoutilizetheamountoftextualdatatoobtainaccurateandcomprehensiveresults.Wepresentamethodologyforbuildingagoodqualityacronymdictionaryofcommonacronymsandtheirexpandedforms,makingeffectiveuseoflargeamountoftexts.ThemethodproposedinthispaperwasappliedtothewholeMEDLINE(7,811,582abstracts).Itextracted886,755candidatesofacronymsandrecognized300,954expandedforms.Detailedevaluationresultsandtheircomparisonwithexistingmethodsarealsogiveninthepaper.

www.1mpi.com

2RELATEDWORK

Acronymrecognitionaimstoextractpairsofshortforms(acronymsorabbreviations)andlongforms(theirexpandedformsorde nitions)occurringintext.Moststudiessharepattern(1)tolocateatextualfragmentwithanacronymanditsexpandedform

Ó2006TheAuthor(s)

This is an Open Access article distributed under the terms of the Creative Commons Attribution Non-Commercial License (www.1mpi.com) which permits unrestricted non-commercial use, distribution, and reproduction in any medium, provided the original work is properly cited.