Indexing support vector machines for efficient top-k classification

This paper proposes an approach to efficiently execute approximate top-k classification (that is, identifying the best k elements of a class) using Support Vector Machines, in web-scale datasets, without significant loss of effectiveness. The novelty of the proposed approach, with respect to other approaches in literature, is that it allows speeding-up several classifiers, each one defined with different kernels and kernel parameters, by using one single index.