Deriving a Primal Form for the Quadratic Power Kernel.

Support vector machines (SVMs) have become a very popular machine learning method for text classification. One reason for their popularity is the possibility of using different types of kernels, which makes them very flexible and allows for dividing data sets that are not linearly separable. This paper shows a method how the primal form of the quadratic power kernel can be derived by using complex numbers. The paper also describes an application of this kernel to text categorization following the Dewey Document Classification (DDC) scheme. In our evaluation, the power kernel (PK) led to a competitive f-measure to those of the compared kernels and was faster to compute than all but the linear kernel.