The encoder-decoder framework achieves state-of-the-art results in keyphrase
generation (KG) tasks by predicting both present keyphrases that appear in the
source document and absent keyphrases that do not. However, relying solely on
the source document can result in generating uncontr