Reproduce “Coreference Graph Guidance for Mind-Map Generation” 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

Zhuowei Zhang, Mengting Hu, Yinhao Bai, Zhen Zhang

Researchers from Nankai University developed the Coreference-Guided Mind-Map Generation Network (CMGN), which integrates explicit coreference graphs and a contrastive learning-based graph enhancement module to improve automatic mind-map generation from text. The method achieved state-of-the-art ROUGE scores, outperforming prior models and even large language models, with a +1.54% increase for salient-sentence-based and +1.24% for key-snippet-based mind-maps.

Key insights

  • Explicitly integrating structural knowledge through coreference graphs and Graph Neural Networks significantly enhances the ability to capture long-range semantic dependencies in mind-map generation.
  • Graph contrastive learning effectively refines and makes more robust the representations derived from potentially noisy external structural cues, leading to higher quality mind-maps.
  • Specialized task-specific architectures that integrate structural information can still outperform general-purpose Large Language Models (LLMs) on complex structured NLP tasks like mind-map generation.

Summary by alphaXiv.

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