Zero-Shot Aspect-Based Scientific Document Summarization using Self-Supervised Pre-training

Amir Soleimani, Vassilina Nikoulina, Benoit Favre, Salah Ait Mokhtar


Abstract
We study the zero-shot setting for the aspect-based scientific document summarization task. Summarizing scientific documents with respect to an aspect can remarkably improve document assistance systems and readers experience. However, existing large-scale datasets contain a limited variety of aspects, causing summarization models to over-fit to a small set of aspects and a specific domain. We establish baseline results in zero-shot performance (over unseen aspects and the presence of domain shift), paraphrasing, leave-one-out, and limited supervised samples experimental setups. We propose a self-supervised pre-training approach to enhance the zero-shot performance. We leverage the PubMed structured abstracts to create a biomedical aspect-based summarization dataset. Experimental results on the PubMed and FacetSum aspect-based datasets show promising performance when the model is pre-trained using unlabelled in-domain data.
Anthology ID:
2022.bionlp-1.5
Volume:
Proceedings of the 21st Workshop on Biomedical Language Processing
Month:
May
Year:
2022
Address:
Dublin, Ireland
Venues:
ACL | BioNLP
SIG:
Publisher:
Association for Computational Linguistics
Note:
Pages:
49–62
Language:
URL:
https://aclanthology.org/2022.bionlp-1.5
DOI:
10.18653/v1/2022.bionlp-1.5
Bibkey:
Cite (ACL):
Amir Soleimani, Vassilina Nikoulina, Benoit Favre, and Salah Ait Mokhtar. 2022. Zero-Shot Aspect-Based Scientific Document Summarization using Self-Supervised Pre-training. In Proceedings of the 21st Workshop on Biomedical Language Processing, pages 49–62, Dublin, Ireland. Association for Computational Linguistics.
Cite (Informal):
Zero-Shot Aspect-Based Scientific Document Summarization using Self-Supervised Pre-training (Soleimani et al., BioNLP 2022)
Copy Citation:
PDF:
https://aclanthology.org/2022.bionlp-1.5.pdf
Code
 asoleimanib/zeroshotaspectbased
Data
FacetSum