论文部分内容阅读
提出了一种修正支撑向量核函数的理论与方法 .与传统的方法相比 ,置换核函数的引入为领域知识与学习模型的融合提供了理论基础与方法论 .该文借助于置换的概念 ,对关于事物模式组成的不变性常识进行了形式化 ,求取了可以定量表述事物模式扰动的置换变换矩阵 ;在分类不变性的约束下 ,运用置换变换矩阵对核函数进行修正 ,获得了改进的学习模型 .文本分类的实验表明 ,学习算法将文本领域内的知识有效地融合到了学习模型中 ,获得了更高的分辨率与泛化能力 .
A kind of theory and method of amending support vector kernel function is proposed.Compared with the traditional method, the introduction of permutation kernel function provides a theoretical basis and methodology for the fusion of domain knowledge and learning model.By means of the concept of permutation, The common sense about the invariance of the patterns of things is formalized and the permutation transformation matrix that can quantitatively describe the perturbations of things is obtained. Under the constraint of classification invariance, the permutation transformation matrix is used to correct the kernel function and the improved learning Model.The experiment of text categorization shows that the learning algorithm effectively integrates the knowledge in the text field into the learning model and achieves higher resolution and generalization ability.