BibTex

Provide the BibTex-formatted reference for the dataset. Please use the correct published version (ACL anthology, etc.) instead of google scholar created Bibtex.

@inproceedings{sun-etal-2021-d2s,
title = "{D}2{S}: Document-to-Slide Generation Via Query-Based Text Summarization",
author = "Sun, Edward  and
Hou, Yufang  and
Wang, Dakuo  and
Zhang, Yunfeng  and
Wang, Nancy X. R.",
booktitle = "Proceedings of the 2021 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies",
month = jun,
year = "2021",
address = "Online",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2021.naacl-main.111",
doi = "10.18653/v1/2021.naacl-main.111",
pages = "1405--1418",
abstract = "Presentations are critical for communication in all areas of our lives, yet the creation of slide decks is often tedious and time-consuming. There has been limited research aiming to automate the document-to-slides generation process and all face a critical challenge: no publicly available dataset for training and benchmarking. In this work, we first contribute a new dataset, SciDuet, consisting of pairs of papers and their corresponding slides decks from recent years{'} NLP and ML conferences (e.g., ACL). Secondly, we present D2S, a novel system that tackles the document-to-slides task with a two-step approach: 1) Use slide titles to retrieve relevant and engaging text, figures, and tables; 2) Summarize the retrieved context into bullet points with long-form question answering. Our evaluation suggests that long-form QA outperforms state-of-the-art summarization baselines on both automated ROUGE metrics and qualitative human evaluation.",
}

Intended Use

What is the intended use of the dataset?

Promote research on the task of document-to-slides generation

Dataset Creators

Who created the original dataset? List the people involved in collecting the dataset and their affiliation(s).

Edward Sun, Yufang Hou, Dakuo Wang, Yunfeng Zhang, Nancy Wang

Funding

Who funded the data creation?

IBM Research

Who added the Dataset to GEM?

Who contributed to the data card and adding the dataset to GEM? List the people+affiliations involved in creating this data card and who helped integrate this dataset into GEM.

Yufang Hou (IBM Research), Dakuo Wang (IBM Research)

How were labels chosen?

How were the labels chosen?

The original papers and slides (both are in PDF format) are carefully processed by a combination of PDF/Image processing tookits. The text contents from multiple slides that correspond to the same slide title are mreged.

Splitting Criteria

Describe any criteria for splitting the data, if used. If there are differences between the splits (e.g., if the training annotations are machine-generated and the dev and test ones are created by humans, or if different numbers of annotators contributed to each example), describe them here.

The dataset integrated into GEM is the ACL portion of the whole dataset described in the paper, It contains the full Dev and Test sets, and a portion of the Train dataset. Note that although we cannot release the whole training dataset due to copyright issues, researchers can still use our released data procurement code to generate the training dataset from the online ICML/NeurIPS anthologies.

Why is the Dataset in GEM?

What does this dataset contribute toward better generation evaluation and why is it part of GEM?

SciDuet is the first publicaly available dataset for the challenging task of document2slides generation, which requires a model has a good ability to "understand" long-form text, choose appropriate content and generate key points.

Proposed Evaluation

List and describe the purpose of the metrics and evaluation methodology (including human evaluation) that the dataset creators used when introducing this task.

Automatical Evaluation Metric: ROUGE Human Evaluation: (Readability, Informativeness, Consistency)

1. Readability: The generated slide content is coherent, concise, and grammatically correct;
2. Informativeness: The generated slide provides sufficient and necessary information that corresponds to the given slide title, regardless of its similarity to the original slide;
3. Consistency: The generated slide content is similar to the original author’s reference slide.

Relevant Previous Results

What are the most relevant previous results for this task/dataset?

Paper "D2S: Document-to-Slide Generation Via Query-Based Text Summarization" reports 20.47, 5.26 and 19.08 for ROUGE-1, ROUGE-2 and ROUGE-L (f-score).

Data Preprocessing

How was the text data pre-processed? (Enter N/A if the text was not pre-processed)

Text on papers was extracted through Grobid. Figures andcaptions were extracted through pdffigures. Text on slides was extracted through IBM Watson Discovery package and OCR by pytesseract. Figures and tables that appear on slides and papers were linked through multiscale template matching by OpenCV. Further dataset cleaning was performed with standard string-based heuristics on sentence building, equation and floating caption removal, and duplicate line deletion.

Filter Criteria

What were the selection criteria?

the slide context text shouldn't contain additional format information such as "*** University"

Consent Policy Details

What was the consent policy?

The original dataset was open-sourced under Apache-2.0.
Some of the original dataset creators are part of the GEM v2 dataset infrastructure team and take care of integrating this dataset into GEM.