Dataset Card for ChemSum

ChemSum Description

• Paper: What are the Desired Characteristics of Calibration Sets? Identifying Correlates on Long Form Scientific Summarization
• Journal: ACL 2023
• Point of Contact: griffin.adams@columbia.edu
• Repository: https://github.com/griff4692/calibrating-summaries

ChemSum Summary

We introduce a dataset with a pure chemistry focus by compiling a list of chemistry academic journals with Open-Access articles. For each journal, we downloaded full-text article PDFs from the Open-Access portion of the journal using available APIs, or scraping this content using Selenium Chrome WebDriver. Each PDF was processed with Grobid via a locally installed client to extract free-text paragraphs with sections.

The table below shows the journals from which Open Access articles were sourced, as well as the number of papers processed.

For all journals, we filtered for papers with the provided topic of Chemistry when papers from other disciplines were also available (e.g. PubMed).

Source	# of Articles
Beilstein	1,829
Chem Cell	546
ChemRxiv	12,231
Chemistry Open	398
Nature Communications Chemistry	572
PubMed Author Manuscript	57,680
PubMed Open Access	29,540
Royal Society of Chemistry (RSC)	9,334
Scientific Reports - Nature	6,826

Languages

English

Dataset Structure

Data Fields

Column	Description
uuid	Unique Identifier for the Example
title	Title of the Article
article_source	Open Source Journal (see above for list)
abstract	Abstract (summary reference)
sections	Full-text sections from the main body of paper (<!> indicates section boundaries)
headers	Corresponding section headers for sections field (<!> delimited)
source_toks	Aggregate number of tokens across sections
target_toks	Number of tokens in the abstract
compression	Ratio of source_toks to target_toks

Please refer to load_chemistry() in https://github.com/griff4692/calibrating-summaries/blob/master/preprocess/preprocess.py for pre-processing as a summarization dataset. The inputs are sections and headers and the targets is the abstract.

Data Splits

Split	Count
train	115,956
validation	1,000
test	2,000

Citation Information

@inproceedings{adams-etal-2023-desired,
    title = "What are the Desired Characteristics of Calibration Sets? Identifying Correlates on Long Form Scientific Summarization",
    author = "Adams, Griffin  and
      Nguyen, Bichlien  and
      Smith, Jake  and
      Xia, Yingce  and
      Xie, Shufang  and
      Ostropolets, Anna  and
      Deb, Budhaditya  and
      Chen, Yuan-Jyue  and
      Naumann, Tristan  and
      Elhadad, No{\'e}mie",
    editor = "Rogers, Anna  and
      Boyd-Graber, Jordan  and
      Okazaki, Naoaki",
    booktitle = "Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)",
    month = jul,
    year = "2023",
    address = "Toronto, Canada",
    publisher = "Association for Computational Linguistics",
    url = "https://aclanthology.org/2023.acl-long.587",
    doi = "10.18653/v1/2023.acl-long.587",
    pages = "10520--10542",
    abstract = "Summarization models often generate text that is poorly calibrated to quality metrics because they are trained to maximize the likelihood of a single reference (MLE). To address this, recent work has added a calibration step, which exposes a model to its own ranked outputs to improve relevance or, in a separate line of work, contrasts positive and negative sets to improve faithfulness. While effective, much of this work has focused on \textit{how} to generate and optimize these sets. Less is known about \textit{why} one setup is more effective than another. In this work, we uncover the underlying characteristics of effective sets. For each training instance, we form a large, diverse pool of candidates and systematically vary the subsets used for calibration fine-tuning. Each selection strategy targets distinct aspects of the sets, such as lexical diversity or the size of the gap between positive and negatives. On three diverse scientific long-form summarization datasets (spanning biomedical, clinical, and chemical domains), we find, among others, that faithfulness calibration is optimal when the negative sets are extractive and more likely to be generated, whereas for relevance calibration, the metric margin between candidates should be maximized and surprise{--}the disagreement between model and metric defined candidate rankings{--}minimized.",
}