Deep pre-trained embedding metrics for evaluation of headline generation
train
andtest
are included in themain.ipynb
notebook- The directory
data
contains human evaluations, angular and ROUGE scores.
FED included in the above notebook. It also contain generated headlines. - Datasets used in our experiments (Amazon product review, CNN/Daily Mail,
Newsroom, Gigaword) are publicly available, thus have not been included in the repo.