Accelerating Spherical k-Means

Spherical k-means is a widely used clustering algorithm for sparse and high-dimensional data such as document vectors. While several improvements accelerations have been introduced the original algorithm, not all easily translate to spherical variant: Many acceleration techniques, algorithms of Elkan Hamerly, rely on triangle inequality Euclidean distances. However, uses Cosine similarities instead distances computational efficiency. In this paper, we incorporate Hamerly working directly with Cosines obtain substantial speedup evaluate these real data.