Hierarchical Learning for Generation with Long Source Sequences

Benchmark Model Rank Results
document-summarization-on-cnn-daily-mailHAT-BART–ROUGE-1: 44.48ROUGE-2: 21.31ROUGE-L: 41.52
reading-comprehension-on-raceHAT (Encoder)–Accuracy: 67.3
text-summarization-on-arxivHAT-BART–ROUGE-1: 46.74ROUGE-2: 19.19ROUGE-L: 42.2
text-summarization-on-pubmedHAT-BART–ROUGE-1: 48.25ROUGE-2: 21.35ROUGE-L: 36.69
text-summarization-on-samsum-corpusHAT-CNNDM–ROUGE-1: 53.01ROUGE-2: 28.27
text-summarization-on-samsum-corpusHAT-CNNDM RL–ROUGE-L: 48.84
text-summarization-on-x-sumHAT-BART–ROUGE-1: 45.92ROUGE-2: 22.79