文本分类中的语义核函数研究

Research on Semantic Kernel Function in Text Classification

传统的很多文本分类算法都是基于文本特征的数值统计信息来进行分类,只考虑特征在文本中的出现频率,而忽略了文本特征之间的语义相关性。针对文本分类任务,本文提出一种基于本体的语义核函数的构造方法,设计和实现了基于WordNet的语义核函数算法,并将该语义核函数嵌入支持向量机分类器中进行文本分类实验,在20NewsGroups数据集上的分类结果表明,基于语义核函数的支持向量机的分类效果明显优于基于线性核的支持向量机的分类效果。

Many traditional text classification algorithms classify documents based on the terms＇ statistical information and they only take into account the frequencies of the terms in indexed documents and in the whole collection but ignore the semantic relevance of the documents＇ terms. In this paper, we propose an approach to the design of a semantic kernel function based on ontology, design and implement an algorithm of WordNet-based semantic kernel function, and then embed this semantic kernel into the Support Vector Machines （SVM） to accomplish a text categorization task. The experimental evaluation on 20 NewsGroups dataset indicates that the performance of the semantic kernel-based SVM outperforms the linear kernel-based SVM.