基于NB的双级分类模型在邮件过滤中的研究

使用朴素的贝叶斯（NB）分类模型对邮件进行分类，是目前基于内容的垃圾邮件过滤方法的研究热点。朴素的贝叶斯在参数之间联系不强的时候分类效果简单而有效。但是朴素的贝叶斯分类模型中对特征参数的条件独立假设无法表达参数之间在语义上的关系，影响分类性能。在朴素的贝叶斯分类模型的基础上，我们提出了一种双级贝叶斯分类模型（DLB，Double Level Bayes），既考虑到了参数之间的影响又保留了朴素的贝叶斯分类模型的优点。同时时DLB模型与朴素的贝叶斯分类模型的性能进行比较。仿真实验表明，DLB分类模型在垃圾邮件过滤应用中的效果在大部分条件下优于朴素的贝叶斯分类模型。

The Research of NB-based DLB Classification Anti-spam

Classification method using Naive Bayesian(NB)classifier model which is the context-based spare filter meth- od,is a hot point.The Naive Bayesian classifier is a simple and effective classification method,but its attribute inde- pendence assumption makes it unable to express its semantic dependence.A new classification model is proposed which we call Double Lever Bayes classifier model(DLB).It considers not only the semantic dependence but also the simple and effective which is the excellence of NB classifier model.The performance is also compared between DLB and NB. The conclusion we get from experiment is that the performance using DLB classifier model is better than which using NB classifier model.