Scholarlink

<p>In the digital age, vast raw data is readily available online. Online news platforms, e-books, research papers, scientific papers, legal information, online blogs, educational websites, and medical papers are among the numerous sources of information flooding the market today. It can be challenging to read all the text available in a short span. Automatic text summarizing (ATS) generates a brief summary while maintaining the context of the written content by applying deep learning methods. Without the help of Artificial Intelligence (AI), preserving context is impossible. The proposed hybrid approach, HPEGPrSumm, is a transformer-based model that employs BERT as an extractive approach and is pre-trained with extracted gap Sensitivity (PEGASUS) as an abstractive approach. This model summarizes the English text, yielding a higher-quality summary through fine-tuning with the aid of the prompt tuning technique, a component of Generative Artificial Intelligence (Gen AI) to preserve the context of the source text. Two established evaluation metrics, Recall Oriented Understudy for Gisting Evaluation (ROUGE) and BERTScore (BS) are carried out to the online news dataset CNN_DailyMail. The proposed and implemented HPEGPrSumm has outperformed the ROUGE and BERTScore with the accuracy percentages in R1, R2, R-L and BS of 50.0, 25.0, 42.9 and 90.5, respectively.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

HPEGPrSumm: a transformer-based text summarization with prompt tuning

  • 期刊论文

  • 发表日期：

    2025年8月12日

查看全文
  • Vaishali,
  • Ginni Arora,
  • Prashant Dixit

摘要

In the digital age, vast raw data is readily available online. Online news platforms, e-books, research papers, scientific papers, legal information, online blogs, educational websites, and medical papers are among the numerous sources of information flooding the market today. It can be challenging to read all the text available in a short span. Automatic text summarizing (ATS) generates a brief summary while maintaining the context of the written content by applying deep learning methods. Without the help of Artificial Intelligence (AI), preserving context is impossible. The proposed hybrid approach, HPEGPrSumm, is a transformer-based model that employs BERT as an extractive approach and is pre-trained with extracted gap Sensitivity (PEGASUS) as an abstractive approach. This model summarizes the English text, yielding a higher-quality summary through fine-tuning with the aid of the prompt tuning technique, a component of Generative Artificial Intelligence (Gen AI) to preserve the context of the source text. Two established evaluation metrics, Recall Oriented Understudy for Gisting Evaluation (ROUGE) and BERTScore (BS) are carried out to the online news dataset CNN_DailyMail. The proposed and implemented HPEGPrSumm has outperformed the ROUGE and BERTScore with the accuracy percentages in R1, R2, R-L and BS of 50.0, 25.0, 42.9 and 90.5, respectively.