一种基于聚类加权的文本特征生成算法  

Text feature generation algorithm based on clustering weighted

作　　者：张凤琴[1] 王磊[1] 张水平[1] 王鹏[2] 程超[1] 

机构地区：[1]空军工程大学信息与导航学院指挥自动化工程系,西安710077 [2]北京航空航天大学软件学院,北京100191

出　　处：《计算机应用研究》2013年第1期146-148,共3页Application Research of Computers

基　　金：陕西省自然科学基金资助项目(2011JM8035)

摘　　要：目前的文本特征生成算法一般采用加权的文本向量空间模型,该模型使用TF-IDF评价函数来计算单个特征的权值,这种算法生成的文本特征冗余度往往都比较高。针对这一问题,采用了一种基于聚类加权的文本特征生成算法,首先对特征候选集进行初始加权处理;然后通过语义和信息熵对特征进行进一步加权处理;最后使用特征聚类对冗余特征进行剔除。实验表明该算法比传统的TF-IDF算法的平均分类准确率高出5%左右。At present,the text feature generation algorithm generally used vector space model（VSM）,the model used TF-IDF evaluation function to calculate individual features weight,the redundancy of text features generated by this algorithm were higher.In order to solve this problem,the paper put forward a text feature generation algorithm based on the clustering of weighted,first the algorithm weighted features with the initial weights,then used the semantics and the entropy to deal with the feature of the further weights,finally used features clustering to undertake eliminate redundant features.Experiments show that the algorithm is better than the traditional TF-IDF algorithm,classification accuracy is higher than the 5% average.

关 键 词：文本分类 特征生成 权值计算 特征聚类 信息熵 

分 类 号：TP391.1[自动化与计算机技术—计算机应用技术]