Abstract: Because Naive Bayes (NB) classification model is simple and effective, good efficiency can be achieved in antispam applications. On the other hand, the assumption of its attribute independence makes it unable to express its semantic dependence. This paper proposed a new antispam classification model based on semi-NB classification model, averaged on N one-dependence classification model. It relaxed the assumption of condition independence of each attribute. It was assumed that all attributes were dependent on one attribute (1-dependence). The average on N 1-dependence was regarded as the probability of each class label. This method is simple and efficient and decreases the classification error ratio.

Key words: Bayesian classification, semi-Naive Bayes, spam

摘要： 由于朴素贝叶斯分类模型的简单高效，在垃圾邮件分类时可以达到较好的效果；但朴素贝叶斯的条件独立假设割裂了属性之间的关系，影响了分类的准确性。放松朴素贝叶斯分类模型关于属性之间条件独立假设，介绍一种新的基于不完全朴素贝叶斯分类模型的垃圾邮件分类模型，N平均1依赖邮件过滤模型。使用N个1依赖分类模型的平均概率作为分类的预测概率。实验证明，该模型在简单、高效的同时降低了对垃圾邮件分类的错误率。

关键词: 贝叶斯分类, 不完全朴素贝叶斯, 垃圾邮件