---
language:
- en
task_categories:
- summarization
pretty_name: SlideTailor-PSP-dataset
---

# SlideTailor-PSP-dataset

[**Paper**](https://huggingface.co/papers/2512.20292) | [**Code**](https://github.com/nusnlp/SlideTailor)

This is the official dataset for the AAAI 2026 paper: **SlideTailor: Personalized Presentation Slide Generation for Scientific Papers**.

This dataset supports research on conditional document-to-slide generation (Paper-to-Slides with Preference). It contains:
- Target research papers (PDF) to be summarized into slides
- Slide templates (PPTX)
- Curated paper–slide pairs for preference extraction
- Configurations for evaluation sets (how to pair inputs/templates/samples)

## Directory structure
- `configs/`: JSON configuration files
- `target_papers/`: target papers (PDFs) and a CSV mapping for metadata
- `slide_templates/`: editable slide deck templates (PPTX), e.g., `1.pptx` … `10.pptx`
- `slide_paper_pairs/`: exemplar pairs used for conditioning
  - `paper/`: paper PDFs
  - `ppt/`: corresponding slide decks

## Configuration format
Each configuration file is a JSON array for one generation settings. Example (`configs/example_config.json`):

```json
[
  {
    "target": "1.pdf",
    "template": "1.pptx",
    "sample": {
      "paper": "1.pdf",
      "ppt": "1.pdf"
    }
  }
]
```

## Citation
```bibtex
@inproceedings{slidetailor,
  title={SlideTailor: Personalized Presentation Slide Generation for Scientific Papers},
  author={Zeng, Wenzheng and Ouyang, Mingyu and Cui, Langyuan and Ng, Hwee Tou},
  booktitle={Proceedings of the AAAI Conference on Artificial Intelligence},
  year={2026}
}
```