Reproduce “D2S: Document-to-Slide Generation Via Query-Based Text 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
Edward Sun, Yufang Hou, Dakuo Wang, Yunfeng Zhang, Nancy X.R. Wang
The D2S system from the University of Michigan and IBM Research automates document-to-slide conversion by treating slide titles as queries for extracting and generating content from source documents, combining information retrieval with abstractive summarization. It also introduces SciDuet, a new dataset of academic paper-slide pairs from NLP and ML conferences, demonstrating performance comparable to human-created slides in readability and informativeness.
Key insights
- Treating slide generation as a query-based single-document summarization task, with the slide title serving as the query, enables more targeted content creation.
- Integrating both information retrieval and abstractive summarization capabilities, guided by keywords from the document's hierarchical structure, produces more relevant and readable slide content than purely extractive methods.
- The availability of a new, aligned dataset (SciDuet) is crucial for training and evaluating models for this specific task, highlighting the need for specialized data.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.