A Rank-Based Approach of Cosine Similarity with Applications in Automatic Classification

This paper introduces a new rank-based approach of the cosine similarity that can be applied in the competitive area of automatic classification. We describe the method and then we present some of its mathematical and computational properties. Tests are performed on a dataset consisting of handwritten digits extracted from a collection of Dutch utility maps. The experimental results are compared with other reported results based on different combining methods which used the same dataset. The obtained results show that the rank-based approach of the cosine similarity can successfully be used for automatic classification or similar tasks.