使用机器学习方法进行新闻情感自动分类.pdfVIP

21 6 V ol . 2 1, No . 6 2007 11 J OU RNA L OF CH INESE IN FO RM AT ION PR OCESSIN G No v., 2007 : 1003- 0077 2007) 06- 0095- 06 , , , 5 18055) : 本文主要研究机器学习方法在新闻文本的情感分类中的应用, 判断其是正面还是负面 们利用朴素贝 叶斯和最大熵方法进行新闻及评论语料的情感分类研究实验表明, 机器学习方法在基于情感的文本分类中也能 取得不错的分类性能, 最高准确率能达到 90 % 同时 们也发现, 对于基于情感的文本分类, 选择具有语义倾向的 词汇作为特征项对否定词正确处理和采用二值作为特征项权重能提高分类的准确率总之, 基于情感的文本分 类是一个更具挑战性的工作 : 计算机应用; 中文信息处理; 文本分类; 情感分析; 贝叶斯; 最大熵 : T P39 1 : A Sentiment Classification for Chinese News Using Machine Learning Methods XU Jun, DING Yu-x in, W AN G Xiao- lon g Intellig ence Computing Research Center , Shenzhen G raduate Schoo l, H arbin Institute T echno log y, Shenzhen , G uang do ng 518055) Abstract: In this paper, w e study how t o apply machine lear ning techniques to so lv e sent iment classif ication prob- lem s . T he main task of sentiment classif icat ion is to determ ine w hether new s or review s is negat ive or posit ive. N a- ive Bay es and M ax imum Entr opy classif ication ar e u sed for the sentiment classif icatio n of Chinese new s and rev iew s . T he ex perimental results show that the methods w e emplo yed perfo rm w ell. T he accur acy of classif icatio n can a- chieve about 90 % . M or eover , w e f ind that selecting the w o rds w ith polarit y as f eatures, neg atio n tagg ing and repre- senting test documents as f eatur e pr esence vector s can impro ve the per formance of sent iment classif ication . Conclu- siv ely , sent iment classif icatio n is a mo re challeng ing pro blem . Key words: computer applicatio n; Chinese informat ion pr ocessing ; tex t catego rization ; sent iment analysis; N aÇv e Ba