Hello you all!
SQuAD dataset creators have used multinomial logistic regression as explained in
this paper and have created 180 million features (sentence detection accuracy for this model was 79%), but it is not clear how they have defined the
target variable or
which are the labels. Because to my mind, as far as we may have two a simple logistic regression would have been sufficient isn't it with an "answer"/"don't answer" feature, isn't it?
If anyone has any idea, please, can you clarify this point ?
Thank you in advance !