Reproduce “NexusSum: Hierarchical LLM Agents for Long-Form Narrative 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
Hyuntak Kim, Byung-Hak Kim
NEXUSSUM is a hierarchical multi-agent Large Language Model framework designed to summarize long-form narratives like books and movie scripts without requiring fine-tuning. The framework achieves state-of-the-art performance, demonstrating up to a 30.0% improvement in BERTScore F1 on the BookSum dataset, while also enabling precise summary length control.
Key insights
- Converting character dialogues to structured third-person descriptions significantly improves input coherence for subsequent summarization.
- Hierarchical processing with distinct LLM agents for preprocessing, summarization, and iterative compression is effective in mitigating information loss over extremely long narrative inputs.
- The multi-agent framework provides fine-tuning-free adaptability, enabling robust performance across diverse narrative types and output requirements through prompt engineering.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.