一种利用近邻和信息熵的主动文本标注方法 An Active Labeling Method for Text Data Based on Nearest Neighbor and Information Entropy.pdf

ISSN 计算机研究与发展 1000—1239／CNII一1777／TP of Researchand JournalComputer Development 一种利用近邻和信息熵的主动文本标注方法 朱 岩 景丽萍 于 剑 (北京交通大学计算机科学系北京100044) (yasmine_zhu@hotmail．corn) AnActive MethodforTextDataBasedonNearest and Labeling Neighbor Information Entropy Zhu Yu Jian Yan，JingLiping，and 100044) (DepartmentofComputerScience，BeijingJiaotongUnirmrsity，Beijing Asitis tolabeltextdocumentsona kindoftext Abstract scale，a quitetime-consuming large a labeleddatais textclassificationand classificationwithfew emerges needed．Thus．semi—supervised fromtraditional textclassification developsrapidly．Different classification，semi—supervisedonly a smallsetof1abeleddataanda setofunlabeleddatatotrainaclassifier．Thesmallset requires large will the oflabeleddataisusedtoinitializetheclassificationmodelinmostcases．Its affect rationality more ofthefinalclassifier．Inordertomakethedistributionofthelabeleddata perfofrmanee consistentwiththedistributionofthe data，fl methodis toavoid original sampling proposed selecting the ofthis theKnearest ofthelabeleddatatobenewcandidatelabeleddata．