Reproduce “SlideTailor: Personalized Presentation Slide Generation for Scientific Papers” 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
Wenzheng Zeng, Mingyu Ouyang, Langyuan Cui, Hwee Tou Ng
The SlideTailor framework introduces preference-guided paper-to-slides generation, distilling implicit user preferences from example presentations and templates to create personalized, editable slides. This approach, which includes a novel "chain-of-speech" mechanism, yields significantly higher quality and user-aligned presentations, demonstrating superior performance on a new benchmark and enabling automated video presentations.
Key insights
- Accurately modeling and leveraging diverse, implicit user preferences, even from unlabeled examples, is crucial for generating truly personalized and high-quality multimodal content like presentation slides.
- An agentic framework combining large language models and vision-language models can effectively distill complex content and aesthetic preferences from natural inputs (e.g., sample presentations and template files) without requiring explicit textual instructions.
- Aligning slide content planning with anticipated oral narration via a "chain-of-speech" mechanism significantly enhances the coherence and clarity of generated slides, directly enabling downstream applications such as automated, speaker-aware video presentations.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.