Top-k similarity search