Relevance feedback in an adaptive space with one-class SVM for content-based music retrieval

In this paper, we develop a novel scheme to content-based music retrieval, using relevance feedback with one-class support vector machine (SVM). Since one-class SVM only concerns the relevant examples and neglects useful information from irrelevant examples provided by the user, an adaptive space is proposed using both relevant and irrelevant examples. The adaptive space, integrated with one-class SVM, transforms the feature space to a space that would better correspond to the userpsilas needs and specificities. Experimental results of retrieval on a music genre database demonstrate the effectiveness of our approach.