改善朴素贝叶斯在文本分类中的稳定性-智能科学网站.pdfVIP

改善朴素贝叶斯在文本分类中的稳定性 石志伟1,2 ， 吴功宜2 （1.天津师范大学计算机与信息工程学院，天津 300074 ；2.南开大学信息技术科学学院，天津300071 ） shi.zw@ 摘要：由于朴素贝叶斯方法具有运行快速、易于实现的特点，它被广泛应用于各种文本分类及信息检索系 统。然而它的不稳定性却一直是一个困扰着研究人员的弱点。本文通过分析朴素贝叶斯的两种常见的实现 模型：二项独立模型（BIM ）和多项模型（MM ），提出混和模型的朴素贝叶斯方法和带有单词量相关的 平滑因子的混和模型。实验表明这两种模型较原来的模型在稳定性方面都有不同程度的提高，尤其是后 者，它在保持朴素贝叶斯快速、易于实现的特点的同时，达到了与线性支持向量机接近的分类性能。 关键字：文本分类；朴素贝叶斯；稳定性 中图法分类号：TP181 文献标识码：A Improving the Stability of Naïve Bayes for Text Categorization Shi Zhi-wei1,2, Wu Gong-yi2 (1.College of Computer and Information Engineering, Tianjin Normal University, Tianjin 300074; 2.Department of Information Science, Nankai University, Tianjin 300071) Abstract ： Naïve Bayes approach is widely used in text categorization and information retrieval, because it is fast and easy to implement. But its instability performance is still a boring drawback. In this paper, we studied two popular models for Naïve Bayes: binary independence model (BIM) and multinomial model (MM), and proposed a hybrid model (HM) and a hybrid model with term-count related smooth factor (HMNSF) for Naïve Bayes. We also performed an experimental evaluation of the effectiveness of HM and HMNSF. The result of our experiments demonstrated that the new models gain better stability than the old ones. And the HMNSF even achieves the competitive performance to the state-of-the-art text classification algorithms such as the Support Vector Machine, while it is still fast and easy to implement. Key words ：text categorization ；Naïve Bayes ；stability 预先定义好的类别当中）在最近十几年中也得到迅 [ ] 1. 引言 速的发展1 ，并且被广泛应用到许多领域中，包