I am just learning about distant supervision. I read the paper of Mintz et al. and trying to get some intuition of how the noise influences the classification.

My general assumption is, that having False-Positives(FP) in our noisy labels hurts the precision of the classifier but not the recall and vice versa False-Negatives(FN) hurt the recall but not the precision.

Q1: Is this assumption correct?

Second, let’s suppose I know that 20% of the data is labeled incorrectly and I know which of these examples are FP or FN.

Q2: Is there a way of estimating the effect this has on measures like Precision or Recall?

I suppose that depends on the classifier but I am not certain.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.