Reproduce “Narrative-Driven Paper-to-Slide Generation via ArcDeck” 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
Tarik Can Ozden, Sachidanand VS, Furkan Horoz, Ozgur Kara, Junho Kim, James Matthew Rehg
ArcDeck, a narrative-driven multi-agent framework from the University of Illinois Urbana-Champaign, generates academic presentations from scientific papers by employing explicit discourse modeling and iterative refinement processes. It produces slide decks with improved narrative coherence and content coverage compared to existing methods, as demonstrated on the new ArcBench evaluation dataset.
Key insights
- Explicit discourse modeling, leveraging Rhetorical Structure Theory, enables the system to capture intricate rhetorical dependencies and build a structured, coherent presentation narrative.
- Establishing and maintaining a "Global Commitment" effectively guides the generation process, ensuring consistency with the paper's overarching thesis and narrative across all slides.
- An iterative multi-agent refinement loop, featuring critique and revision, is vital for achieving fine-grained structural coherence and high-quality presentation outlines.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.