基于U形有向无环支持向量机的多类文本分类研究

针对原始的有向无环支持向量机（Directed Acyclic Graph Support Vector Machine，DAGSVM）多类文本分类算法结构图中存在误差累积以及路径冗余的问题，提出一种改进的有向无环图支持向量机多类文本分类算法。将易于分离的那些类两两组成两类分类器放置在结构图的上层，有效克服了误差累积问题，通过将原始的分类结构图改造为U形图以及引入集合运算进行类别判定，有效剪除冗余路径并降低时间复杂度。实例表明，该算法能有效提高类别判定的准确率并降低时间复杂度。

An Improved SVM Multiclass Classification Algorithm Based on U-shaped DAG

For the multiclass text classification algorithm based on original DAGSVM （Directed Acyclic Graph Support Vector Machine） has error accumulation and path redundancy problems in its structure diagram, this paper proposes an improved multiclass text categorization method based on DAGSVM. In this approach, binary class text classification consisting of two classes from those easy separating classes is placed in the upper layer of the structure diagram of DAGSVM, and it effectively overcome error accumulation problem. The reconstructing the original structure diagram into U -shaped diagram, as well as the introducing set operations into classification, is effective trims the redundant paths and reduces time complexity. The experimental results show this approach can effectively improve the accuracy of classification and significantly reduce the time complexity.