Reproduce “OrderSum: Semantic Sentence Ordering for Extractive Summarization” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Taewan Kwon, Sangyong Lee

OrderSum presents an extractive summarization model designed to semantically order selected sentences, thereby enhancing summary coherence beyond mere content selection. The model establishes new state-of-the-art ROUGE-L scores, notably surpassing previous approaches by 2.54 points on the CNN/DailyMail dataset.

Key insights

  • The semantic order of sentences is as crucial as content selection for generating high-quality, coherent extractive summaries.
  • Explicitly optimizing for sentence order through a summary-level reranking framework leads to more readable and semantically aligned summaries.
  • Integrating full-text ROUGE-L into a contrastive learning objective can effectively train models to prioritize structural and ordering similarity to a reference summary.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.