Reproduce “RTSUM: Relation Triple-based Interpretable Summarization with Multi-level Salience Visualization” 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
Seonglae Cho, Yonggi Cho, HoonJae Lee, Myungha Jang, Jinyoung Yeo, Dongha Lee
Researchers from Yonsei University developed RTSUM, an unsupervised, interpretable summarization framework that extracts salient relation triples from text and transforms them into fluent summaries. The framework provides multi-level salience visualization, allowing users to understand content selection at the sentence, relation, and phrase levels.
Key insights
- Combining fine-grained extractive units (relation triples) with abstractive generation offers a robust hybrid approach to summarization, balancing precision and fluency.
- Multi-level salience scoring, incorporating sentence, relation triple, and phrase importance, provides a comprehensive understanding of textual significance for information selection.
- Self-supervised training with careful data filtering can effectively teach a language model to "sentencify" structured information, circumventing the need for expensive human-annotated summaries.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.