一种基于同义词扩展的不平衡文本分类方法  

An Imbalanced Text Classification Method Based on Synonyms Expansion

作　　者：杨鸿骏[1,2] 周亚建[1,2] 郭玉翠[3] 

机构地区：[1]北京邮电大学信息安全中心,北京100876 [2]灾备技术国家工程实验室,北京100876 [3]北京邮电大学理学院,北京100876

出　　处：《情报杂志》2013年第9期204-206,F0003,共4页Journal of Intelligence

基　　金：国家自然科学基金项目"基于行为分析的网络流量检测技术研究"(编号:60972077)的资助

摘　　要：针对传统文本分类方法的性能,尤其是其中少数类的分类性能会随着文本不平衡程度的加重而迅速恶化的现象,提出了一种基于同义词扩展的不平衡文本分类改进方法。该方法通过建立同义词词典、确定扩展规则和调整"特征保持因子"等几个步骤,实现了少数类中的特征项的丰富和补偿,同时对扩展带来的原文档特征变化予以了补偿。实验结果表明,该方法可以从很大程度上改善少数类的分类性能,并且随着少数类中文本数量的减少,性能的提升会越发显著。与此同时,分类器的总体分类性能也得到了一定程度的提升。The performance of traditional text categorization methods, especially the categorization performance for minority classes, often deteriorates rapidly for imbalanced text. A new method based on synonyms expansion is introduced in this paper in order to deal with im- balanced text classification. With the steps of the establishment of synonym-dictionary, the determination of expansion rules and the modi- fication of the ＂Feature-Maintaining Factor＂, feature items of minority classes are enriched. At the same time, the changes brought by the expansion are compensated. The experimental results show the categorization performance for minority classes is improved to a high de- gree. Moreover, with the decrease of the quantity of text in minority classes, the performance improves significantly. The overall perform- ance is improved to some degree at the same time.

关 键 词：文本分类 不平衡数据集 同义词词典 词频保持 

分 类 号：TP391.1[自动化与计算机技术—计算机应用技术]