摘要:
垃圾邮件过滤是信息时代的一个重要研究课题,一封重要邮件被错分会产生不可估量的代价.因此,如何提高过滤器的性能成为垃圾邮件过滤领域中的核心问题.目前,业界通常采用机器学习算法中的二分类模型以处理垃圾邮件过滤问题.然而,较之于三支决策模型,二分类模型会产生较大的错分代价.作为三支决策的一个重要分支,基于决策理论粗糙集的三支决策模型符合人类认知习惯,且能有效地降低错分代价,进而提高过滤器的性能.然而,在构造损失函数时,少有研究考虑由于等价类之间的差异性而对分类结果带来的影响.因此,在基于决策理论粗糙集的三支决策模型的基础上,提出了一种基于相似度量的自适应三支垃圾邮件分类模型.该模型根据集合方差计算了条件属性的权重,并基于相似度量建立了一种刻画差异信息的综合评价函数,进而根据贝叶斯决策规则构建了一种计算自适应阈值对的方法.实验结果表明所提模型在垃圾邮件过滤领域表现优异.
Abstract:
Spam filtering is an important issue in the information age. And, if an important email is wrongly classified, it would lead to an immeasurable cost. Thus, in the field of spam filtering, the ways to improve the accuracy and recall of the filters is the key issue. At present, the binary classification model in machine learning is usually used to deal with spam filtering. However, compared with the three-way decisions, the binary classification model usually leads to a higher cost of misclassification. And, as an important branch of three-way decisions, the three-way decisions with decision-theoretic rough sets can effectively reduce the misclassification cost and further improve the performance of filters. And, it also conforms to human cognition. Nevertheless, few studies consider the effect on classification results induced by the differences among equivalence classes when constructing the loss functions. Therefore, under the framework of the three-way decisions with decision-theoretic rough sets, an adaptive three-way spam filter with similarity measure is proposed. The model calculates the weights of condition attributes according to set variance firstly. Then, a comprehensive evaluation function for describing difference information among equivalence classes based on similarity measure of set is established. Finally, an adaptive model for calculating threshold pairs based on Bayesian decision rules is constructed. Experimental results show that the proposed model performs well in the field of spam filtering.
百度学术
下载:
