summarisation
scientific_lay_summarisationThis repository contains the PLOS and eLife datasets, introduced in the EMNLP 2022 paper "[Making Science Simple: Corpora for the Lay Summarisation of Scientific Literature
](https://arxiv.org/abs/2210.09932)".
Each dataset contains full biomedical research articles paired with expert-written lay summaries (i.e., non-technical summaries). PLOS articles are derived from various journals published by [the Public Library of Science (PLOS)](https://plos.org/), whereas eLife articles are derived from the [eLife](https://elifesciences.org/) journal. More details/anlaysis on the content of each dataset are provided in the paper.
Both "elife" and "plos" have 6 features:
- "article": the body of the document (including the abstract), sections seperated by "/n".
- "section_headings": the title of each section, seperated by "/n".
- "keywords": keywords describing the topic of the article, seperated by "/n".
- "title" : the title of the article.
- "year" : the year the article was published.
- "summary": the lay summary of the document.openai-tldr-summarisation-preferences
Human feedback data
This is the version of the dataset used in https://arxiv.org/abs/2310.06452.
If starting a new project we would recommend using https://huggingface.co/datasets/openai/summarize_from_feedback.
See https://github.com/openai/summarize-from-feedback for original details of the dataset.
Here the data is formatted to enable huggingface transformers sequence classification models to be trained as reward functions.
scientific_lay_summarisation-plos-norm
scientific_lay_summarisation - PLOS - normalized
This dataset is a modified version of tomasg25/scientific_lay_summarization and contains scientific lay summaries that have been preprocessed with this code. The preprocessing includes fixing punctuation and whitespace problems, and calculating the token length of each text sample using a tokenizer from the T5 model.
Original dataset details:
Repository: https://github.com/TGoldsack1/Corpora_for_Lay_Summarisation
Paper: Making… See the full description on the dataset page: https://huggingface.co/datasets/pszemraj/scientific_lay_summarisation-plos-norm.DPO-tldr-summarisation-preferences
Dataset Card for DPO-tldr-summarisation-preferences
Reformatted from openai/summarize_from_feedback dataset.
The LION-series are trained using an empirically optimized pipeline that consists of three stages: SFT, DPO, and online preference learning (online DPO). We find simple techniques such as sequence packing, loss masking in SFT, increasing the preference dataset size in DPO, and online DPO training can significantly improve the performance of language models. Our best models… See the full description on the dataset page: https://huggingface.co/datasets/Columbia-NLP/DPO-tldr-summarisation-preferences.scientific_lay_summarisation-elife-norm
scientific_lay_summarisation - elife - normalized
This is the "elife" split. For more words, refer to the PLOS split README
Contents
load with datasets:
from datasets import load_dataset
# If the dataset is gated/private, make sure you have run huggingface-cli login
dataset = load_dataset("pszemraj/scientific_lay_summarisation-elife-norm")
dataset
Output:
DatasetDict({
train: Dataset({
features: ['article', 'summary', 'section_headings', 'keywords', 'year'… See the full description on the dataset page: https://huggingface.co/datasets/pszemraj/scientific_lay_summarisation-elife-norm.pmc-medical-summarisation
