Rather than calculated and averaged over each document instance like in the example-based measures, label-based measures are calculated and averaged over each category. In each of the one-versus-all binary classification step, we can create the matrix in Table 1 for the corresponding category c. Then for this one category c,