【作者】 张美香；

【导师】 陈俊杰；

【作者基本信息】 太原理工大学 ， 计算机应用技术， 2005， 硕士

【摘要】 随着网络信息的发展,信息自动分类已经成为人们获取有用信息不可或缺的工具。贝叶斯作为其中的一种分类方式,应用在众多领域。贝叶斯方法的一大优点是利用了先验信息,能够在不确定性的推理中提供一种模式和处理方法。 本文首先对文本分类系统以及贝叶斯分类模型作了分析和探讨,包括文本信息的表示、提取,文本分类的方法以及贝叶斯用于文本分类的模型和算法。然后,本文分析了贝叶斯分类的数据稀疏的问题,讨论了所采用的laplace平滑方法的缺陷,提出了用统计语言模型uni-gram的平滑方法来改进数据稀疏状况,并介绍了uni-gram模型的三种平滑方法,分别是Jelinek-Mercer平滑方法、Dirichlet方法以及绝对折扣法。 最主要的工作是用统计语言模型的平滑方法改进了贝叶斯分类器,就是用uni-gram的三种平滑方式代替了贝叶斯原来的laplace平滑,提出了具体的算法和实现框图。并且对改进了的贝叶斯进行了实验分析,选择了合适的平滑参数取值,与原来的分类器在性能上作了比较,取得了较高的分类准确率和召回率。 今后,应该用统计语言模型的Bi-gram,Tri-gram模型来更好更多还原

【Abstract】 Along with the development of network information, automatic information classification has been an essential tool to gain useftil information. As a classification method, Naive Bayes classifier has been applied to many fields. The advantage of Naive Bayes method lies in the usage of prior information, which could provide a pattern and management method under the incertitude logic.First of all, this paper described text classification system, the content includes text information expressing、 extracting and themethod of text classification. Subsequently article discussed Bayes classifier model and algorithm.And then, this paper introduced data sparse condition of Bayes classifier. Disadvantage of laplace smoothing used by traditional Naive Bayes classifier has been pointed out, Another smoothing method is advised to replace Laplace. The advised method just is statistical language model(uni-gram) smoothing: Jelinek-Mercersmoothing , Dirichlet smoothing and Absolute-discounting smoothing.Mostly, we have improved Bayes classifier with new smoothing methods of statistical language model, that is to say, we replaced laplace smoothing of traditional Bayes classifier with other three smoothing methods of uni-gram model. Specific algorithm and framework have been shown. In order to affirm our work, we test three uni-gram smoothing used in Bayes classifier, select suitable parameter value. Our experimental results show that using a language model, we are able to obtain better performance than traditional Naive Bayes classifier.In the future, we should improve Bayes classifier with statistical language model (Bi-gram model and Tri-gram model) smoothing.更多还原