UNIST-Eunchan/NLP-Paper-to-QA-Generation

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

<!-- Provide a longer summary of what this dataset is. -->

• —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

<!-- Provide the basic links for the dataset. -->

• —Repository: allenai/qasper

Uses

• —Question Generation from Research Paper
• —Long-Document Summarization
• —Question-based Summarization <!-- Address questions around how the dataset is intended to be used. -->

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.