摘要
Namedentityrecognitionisafundamentaltaskinbiomedicaldatamining.Inthisletter,anamedentityrecognitionsystembasedonCRFs(ConditionalRandomFields)forbiomedicaltextsispresented.Thesystemmakesextensiveuseofadiversesetoffeatures,includinglocalfeatures,fulltextfeaturesandexternalresourcefeatures.Allfeaturesincorporatedinthissystemaredescribedindetail,andtheimpactsofdifferentfeaturesetsontheperformanceofthesystemareevaluated.Inordertoimprovetheperformanceofsystem,post-processingmodulesareexploitedtodealwiththeabbrevia-tionphenomena,cascadednamedentityandboundaryerrorsidentification.Evaluationonthissystemprovedthatthefeatureselectionhasimportantimpactonthesystemperformance,andthepost-processingexploredhasanimportantcontributiononsystemperformancetoachievebetterre-sults.
出版日期
2007年06月16日(中国期刊网平台首次上网日期,不代表论文的发表时间)