基于NaiveBayes的CLIF_NB文本分类学习方法_刘丽珍.pdfVIP

26 9 Vol. 26 No. 9 2005 9 MINI- MICRO SY ST EMS Sep. 2005 Naive Bayes CLIF- NB 1, 2 2 3 刘丽珍 , 宋瀚涛, 陆玉昌 1( , 100037) 2( , 100081) 3( , 100084) E-mail: liulizh en@ bit . edu. n : 针对Naive Bayes 方法中条件独立性假设常常与实际相违背的情况, 提出了CLIF NB 文本分类学习方法, 用互信息 - 理论, 计算特征属性之间的最大相关性概率, 用变量集组合替代线性不可分属性, 改善条件独立性假设的限制, 并通过学习一系 列分类器, 缩小训练集中的分类错误, 综合得出分类准确率较高的CLIF NB 分类器. - : ; Naive Bay es; : T P 393: A : 1000-1220( 2005) 09-1575-03 Method of CLIF- NB Text Classification Learning Based on Naive Bayes 1,2 2 3 - , - , - LIU Li zhen SONG Han tao LU Yu hang 1( , , 100037, , ) Inf ormation Engineering College Cap ital Normal University Beij ing China 2( Dep artment of Comp uter, Beji ing I nstitute of Technology , Beij ing 100081, China) 3( , , 100084, ) Dep artment of Comp uter Tsinghua University Beij ing China : . Abstract The method of CLIF- NB t ext lassifi ation learning based on Naive Bay es is proposed T o solve the problem that in- dependen e hypothesis is not oin ident with the a tual situation in Naive Bayes lassifi ation method, the paper uses the theory , , of m ut ual informat ion and al ulate the max imum relative probability during training the text feature propert ies and import variables set t o ombine and repla e line inseparable attribut es. So the method an improve the limit of onditional indepen-