Reproduce “Sentence Centrality Revisited for Unsupervised 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
Hao Zheng, Mirella Lapata
Researchers at the University of Edinburgh introduced PACSUM, an unsupervised extractive summarization approach that leverages fine-tuned BERT embeddings and position-augmented directed graphs to determine sentence centrality. This method achieved ROUGE F1 scores competitive with leading supervised summarization systems across English and Chinese news datasets, without requiring large-scale labeled training data.
Key insights
- Contextualized sentence embeddings, derived from a specially fine-tuned BERT model, significantly enhance sentence similarity calculations for unsupervised summarization.
- Incorporating directionality and position-based differential weighting into graph edges dramatically improves salience detection compared to traditional undirected graph methods.
- Unsupervised summarization, when leveraging advanced neural representations and refined graph structures, can achieve performance levels comparable to state-of-the-art supervised models on various tasks and languages.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.