dataset_info:
features:
- name: 'Unnamed: 0'
dtype: int64
- name: question
dtype: string
- name: answer
dtype: string
- name: abstract
dtype: string
- name: introduction
dtype: string
splits:
- name: train
num_bytes: 1844987
num_examples: 421
- name: validation
num_bytes: 949747
num_examples: 211
- name: test
num_bytes: 1403003
num_examples: 320
download_size: 2341682
dataset_size: 4197737
configs:
- config_name: default
data_files:
- split: train
path: data/train-*
- split: validation
path: data/validation-*
- split: test
path: data/test-*
license: mit
task_categories:
- summarization
- question-answering
language:
- en
tags:
- nlp-research-paper-abstract
- nlp-research-paper
- question-generation
pretty_name: NLP_Papers_to_Question_Generation
size_categories:
- n<1K
Dataset Card for Dataset Name
This dataset was created by modifying and adapting the allenai/QASPER: a dataset for question answering on scientific research papers dataset and aims to generate Question-Answer Pairs from the Abstract, Introduction of an NLP Paper.
Dataset Description
First, we extracted the abstract, introduction of each NLP paper from QASPER dataset.
We also extracted only the rows labeled question and answer that had an abstract answer rather than extractive.
train : 421 rows
validation : 211 rows
test : 320 rows
Curated by: @UNIST-Eunchan
Dataset Sources
This data is made by applying and processing allenai/qasper
- Repository: allenai/qasper
Uses
- Question Generation from Research Paper
- Long-Document Summarization
- Question-based Summarization
Dataset Creation
Curation Rationale
Long Document Summarization datasets, especially those for Research Paper Summarization, are very limited and scarce.
We tweak the existing data to provide domains and QA pairs specific to NLP among Research Papers.
We expect to be able to generate multiple QA pairs if we let the model sample through training.
We will release the fine-tuned model in the future.