Example high confidence positive / high confidence negative sentences
cgreene opened this issue · comments
This came up in meetings around this work a number of times. It would be good to produce, for each edge type, a table with high confidence positive (near 1) and high confidence negative (near 0) edges. You should definitely produce this for the discriminative model that you describe in the calibration section. It would probably also be good to produce this for the corresponding generative model.