Abstract: Along with the technical development of network and information technology，the text categorization becomes the key technique on processing and organizing large scale of texts.How to characterize the text exactly as a data set that can be processed is a key problem that restricts the improvement of text categorization results seriously.The author brings up a formula used to characterize the text named p-idf based on the vector space model and tf-idf.After comparing Bayes，K neighbors，neural network and support vector machine these four typical text categorization devices，the author builds a text categorization system using support vector machine.After a scientifical test which displays the impact to the text categorization system of these three formula tf-idf，p-idf，LTC，we can conclude that the p-idf formula is reasonable and valid to a text categorization system.

Key words: text categorization, vector space model, p-idf, Support Vector Machine（SVM）

摘要： 随着网络和信息技术的迅猛发展，文本分类成为处理和组织大量文档数据的关键技术。文本的特征表示严重地限制了文本分类性能的提升。以经典的向量空间模型和tf-idf权值计算公式为基础，提出了以应用于文本分类为目的的权值改进公式p-idf公式。在比较了贝叶斯、K近邻、神经网络和支持向量机四种典型的文本分类器的基础上，采用支持向量机分类器搭建了一个文本分类试验系统。经过科学的试验比较了tf-idf、p-idf、LTC三种权值公式在文本分类系统中对分类器性能的影响，证实了所提出的p-idf公式的合理性和有效性。

关键词: 文本分类, 向量空间模型, p-idf, 支持向量机