Interest Extraction Using Relevance Feedback with Kernel Method

Abstract

In this paper, we propose interest extraction using the relevance feedback with the kernel method. In the field of machine learning, the kernel method has been used. Since the classifier using the kernel method creates a discriminant function in a feature space, the discriminant function is a nonlinear function in a input space. The kernel method is used for the Support Vector Machine (SVM), the Kernel PCA, and so on. The SVM set a discriminant hyperplane between positive data and negative data. Hence, a distance between the hyperplane and a training sample is not important in the SVM. It is difficult to use the SVM to score other samples. Our goal is to create a method which scores the other samples in the feature space. We propose the relevance feedback which is carried out in the feature space. Hence, this relevance feedback can deal with nonlinearity of data. We compare the proposed method with the common relevance feedback using test collection NTCIR2. Finally, we comfirm the proposed method is superior to the common method through simulations.