Can I read SeburSum: a novel set-based summary ranking strategy for summary-level extractive summarization on EtoBox?

SeburSum: a novel set-based summary ranking strategy for summary-level extractive summarization by Shuai Gong; Zhenfang Zhu; Jiangtao Qi; Wenqing Wu; Chunling Tong is a Computer Science article available to read on EtoBox.

What is SeburSum: a novel set-based summary ranking strategy for summary-level extractive summarization about?

level extractive summarization is often regarded as a text-matching task, which selects the summary that is semantically closest to the source document by a matching model. However, the method tends to select candidate summaries with more sentences, because it calculates the semantic similarity between the candidate summaries and the source document. To address the issue, we propose a novel set-based summary ranking strategy for extractive summarization called Sebur-Sum, which selects a summary by examining its semantic similarity with mutually exclusive candidate summaries, rather than its similarity to the source document. In contrast to conventional extractive summarization methods that rely on well-trained extractive models trained on labeled data to select sentences, the ranking strategy eliminates the need for labeled data and enhances the versatility of both supervised and unsupervised extractive summarization. To improve the calculation accuracy of semantic similarity between candidates, we construct a contrastive learning framework with a task-specific contrastive loss to learn vector representations for each candidate summary. Experimental results show that in the supervi

Who reads SeburSum: a novel set-based summary ranking strategy for summary-level extractive summarization?

Author

Shuai Gong; Zhenfang Zhu; Jiangtao Qi; Wenqing Wu; Chunling Tong

Publisher

Springer Science and Business Media LLC

Published

2023

Language

EN

Field

Computer Science (Physical Sciences)