Reproduce “MRGSEM-Sum: An Unsupervised Multi-document Summarization Framework based on Multi-Relational Graphs and Structural Entropy Minimization” 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
Yongbing Zhang, Fang Nan, Shengxiang Gao, Yuxin Huang, Kaiwen Tan, Zhengtao Yu
MRGSEM-Sum, an unsupervised multi-document summarization framework, leverages multi-relational graphs and 2D structural entropy minimization to adaptively cluster sentences and extract summaries. It consistently outperforms unsupervised baselines and achieves performance comparable to supervised models and large language models across diverse datasets, including long documents and non-news domains.
Key insights
- Integrating diverse relationship types (semantic and discourse) into a multi-relational graph provides a more comprehensive and accurate model of inter-sentence connections for summarization.
- Applying 2D structural entropy minimization offers an effective, parameter-free approach to adaptively determine cluster structures in graphs, overcoming the limitation of pre-specifying cluster numbers.
- Incorporating positional importance during summary extraction significantly enhances summary quality and salience, particularly for longer documents where key information often appears early.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.