SEQUENCE-TO-SEQUENCE DEEP NEURAL MODELS FINE-TUNED

Copied to clipboard

Challenge:	Existing methods for abstractive summarization are limited by the quality of the first-stage candidates.
Approach:	They propose a method that fuses several summary candidates to produce a novel abstractive second-stage summary.
Outcome:	The proposed method improves ROUGE scores and qualitative properties of fused summaries on several summarization datasets.

Papers with Sequence-to-sequence deep neural models fine-tuned