mine model linkag inform citat context improv biomed literatur retriev brain informat proceed acm confer inform knowledg manag brain informat revisit rocchio relev feedback algorithm probabilist model explor social annot tag enhanc inform retriev perform enhanc content dbase imag retriev machin learn techniqu optim ir promot rank divers biomed inform retriev wikipedia medic search classif tool recommend dplasa adapt sentiment analysi applic sale perform predict surviv model approach biomed search result diversif wikipedia qualiti dawar model sale predict review passag extract result combin genom inform retriev rank biomed inform retriev context integr multipl document featur languag model expert find rank biomed inform retriev context bioclink probabilist approach improv genom search citat link discov tempor associ signific gene express diverg pattern discov signific frequenc chang dissimilar larg databas boost biomed inform retriev perform citat graph empir studi bayesian learn approach promot divers rank biomed inform retriev graph dbase approach mine multilingu word associ wikipedia trec dchem larg scale chemic inform retriev evalu trec model document featur expert find integr genom machin learn approach inform retriev applic web model predict help onlin review dynam window base passag extract algorithm genom inform retriev rerank model genom aspect search york univers trec blog track helpmet nonlinear model predict help onlin review york univers trec enterpris document search york univers trec genom track knowledg retriev kr life lost live wisdom lost knowledg knowledg lost inform eliot rock increas growth data inform find knowledg real challeng urgent task tradit data inform retriev system support current web longer adequ knowledg seek task knowledg retriev system gener retriev system serv purpos basic issu knowledg retriev system examin conceptu framework system propos theori technolog theori knowledg machin learn knowledg discoveri psycholog logic infer linguist briefli mention implement knowledg retriev system applic knowledg retriev rough set biomed domain present comparison interesting measur web usag mine empir studi find molecular complex multipl layer cluster protein interact network comparison interesting measur web usag mine empir studi effect extens okapi biomed text mine multilevel search drank framework inform retriev appli data mine pseudo drelev feedback high perform text retriev paper investig data mine text classif train techniqu identifi relev passag base small set label passag obtain blind feedback retriev system data mine result expand queri term estim paramet probabilist weight function evalu data mine base feedback method trec hard data set result show data mine successfulli appli improv text retriev perform report experiment find detail mine model databas user access pattern boost predict accuraci imbalanc dataset svm ensembl abstract learn imbalanc dataset inher difficult due lack inform minor class paper studi perform svm gain great success real applic imbalanc data context empir analysi show svm suffer bias decis boundari predict perform drop dramat data highli skew propos combin integr sampl techniqu ensembl svm improv predict perform integr sampl techniqu combin sampl sampl techniqu empir studi show method outperform individu svm state art classifi boost predict accuraci imbalanc dataset svm ensembl abstract learn imbalanc dataset inher difficult due lack inform minor class paper studi perform svm gain great success real applic imbalanc data context empir analysi show svm suffer bias decis boundari predict perform drop dramat data highli skew propos combin integr sampl techniqu ensembl svm improv predict perform integr sampl techniqu combin sampl sampl techniqu empir studi show method outperform individu svm state art classifi concept dbase biomed text retriev platform okapi dbase contextu inform retriev york univers trec enterpris email discuss search york univers trec genom track york univers trec legal track york univers particip legal track year track develop okapi base legal search engin lse experi focus evalu effect probabilist text retriev model legal domain order address special problem legal text retriev automat feedback method term weight method propos test hierarch group associ rule applic real dworld domain appli languag model session identif databas trace log find analyz databas user session machin learn approach identifi databas session unlabel data contextu inform improv retriev perform distanc dbase algorithm cluster databas user session dual index model contextu inform retriev york univers trec genom track genom experi focu address major problem biomed inform retriev problem deal synonym deal frequent acronym deal homonym propos queri expans algorithm construct structur queri experi averag precis map automat run york ga algorithm manual run york gm algorithm evalu result show algorithm ectiv improv retriev perform find techniqu pseudo relev feedback extend stop word set make posit contribut retriev perform york univers trec spam track propos variant nearest neighbor classif method call instanc weight nearest neighbor method adapt spam filter method assign weight distanc weight correct weight train instanc make weight classifi email correct weight mainten train data make train data adapt spam characterist submit spam filter spam track filter pure base instanc weight knn method filter combin knn method spam filter classif techniqu report offici result submiss spam track evalu data set york univers trec hard track york univers trec spam track propos variant nearest neighbor classif method call instanc weight nearest neighbor method adapt spam filter method assign weight distanc weight correct weight train instanc make weight classifi email correct weight mainten train data make train data adapt spam characterist submit spam filter spam track filter pure base instanc weight knn method filter combin knn method spam filter classif techniqu report offici result submiss spam track evalu data set york univers trec terabyt track york univers trec hard genom track cluster web surfer probabilist model real applic dynam web log session identif statist languag model present session identif method base statist languag model featur select rough set web page classif abstract web page classif problem assign predefin categori web page challeng web page classif deal high dimension featur space present featur reduct method base rough set theori investig effect rough set featur select method web page classif experi rough set featur select improv predict perform origin featur set repres web page larg object subject algorithm group associ rule propos algorithm group summar associ rule algorithm recurs group rule structur rule gener tree cluster result algorithm group rule semant distanc rule make automet tag semant tree structur network item provid case studi propos algorithm evalu result show group method effect produc good group result text classif asian languag word segment present simpl approach asian languag text classif word segment base statist gram languag model examin chines japanes text classif appli machin learn text segment inform retriev propos supervis word segment techniqu text segment chines inform retriev method combin advantag tradit dictionari base charact base mutual inform base approach overcom shortcom experi trec data show method promis method complet languag independ unsupervis promis avenu construct accur multi lingual cross lingual inform retriev system exibl adapt segment accuraci supervis segment high segment method give compar case retriev perform commonli believ word segment accuraci monoton relat retriev perform chines inform retriev chines relationship segment retriev perform fact nonmonoton word segment accuraci segment phenomenon begin occur lead reduct inform retriev perform demonstr eect present empir investig inform retriev chines trec data wide varieti word segment algorithm word segment accuraci rang includ word segment accuraci supervis word segment approach organ knowledg polit process reconfigur context bpr comparison interesting function learn web usag pattern investig relationship word segment perform retriev perform chines ir commonli believ word segment accuraci monoton relat retriev perform chines inform retriev paper show chines relationship segment retriev perform fact nonmonoton word segment accuraci segment phenomenon begin occur lead reduct inform retriev perform demonstr ect present empir investig inform retriev chines trec data wide varieti word segment algorithm word segment accuraci rang appear main reason drop retriev perform correct compound colloc preserv accur segment broken accur reason segment surpris advantag suggest word broad notion conveni captur gener semant mean chines text discoveri interest associ rule livelink web log data dsupervis word segment chines inform retriev propos supervis word segment techniqu chines inform retriev method combin advantag tradit dictionari base approach charact base approach overcom shortcom experi trec data show compar perform dictionari base charact base approach method languag independ unsupervis promis avenu construct accur multilingu inform retriev system flexibl adapt probabl dbase chines text process retriev case studi learn imbalanc data set probabl dbase chines text process retriev okapi chines text retriev experi trec introduct focu okapi trec chines experi investig ectiv di erent automat index method phrase weight retriev base probabilist model chines text compar di erent probabilist weight method base rang word singl charact approach index method experi index method linguist unit word compound word phrase text index term repres text refer method word approach approach text segment divid text linguist unit regard precursor bottleneck kind system method index text base singl chines charact text index bythecharact appear text singl charact approach search conduct anymulti charact word okapi trec introduct summari gener citi submit run automat ad hoc larg collect track automat rout chines track part interact ltere track signi develop okapi style weight trec trec time round attempt ad hoc notabl chines experi extend weight cover search word phrase submit run chines incorpor run time passag determin search okapi back end search engin consider speed function incorpor section automat ad hoc vlc track good deal fruitless explor includ work term phrase method nalli trec expans term top document retriev 