Scholarlink

<p>Question Generation is a research area focused on the automatic creation of relevant questions from given texts, providing critical support across various industries, including education and healthcare. Achieving high accuracy and diversity in the questions generated remains a considerable challenge. In this paper, we introduce a Diverse Question Generation (DQG) framework that aims to generate a wide array of questions based on distinct selections. This framework leverages additional information, such as question types and attributes, to ensure the generation of specific questions. Furthermore, we incorporate a data augmentation method to enhance the performance of the generated questions. This approach employs large language models with a designed prompt template to generate additional samples that are relevant to the training dataset. We train and evaluate our model on the three datasets designed for question generation. Our experimental results demonstrate a significant performance improvement compared to the original FLAN-T5 models.</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

Diversity Question Generation with Desired Information Selection and Training Data Augment

  • 期刊论文

  • 发表日期：

    2025年4月7日

查看全文
  • Kunxiao Liu,
  • Shijin Zhang,
  • Yuqiang Wu,
  • Xi Wu,
  • Lyulog He

摘要

Question Generation is a research area focused on the automatic creation of relevant questions from given texts, providing critical support across various industries, including education and healthcare. Achieving high accuracy and diversity in the questions generated remains a considerable challenge. In this paper, we introduce a Diverse Question Generation (DQG) framework that aims to generate a wide array of questions based on distinct selections. This framework leverages additional information, such as question types and attributes, to ensure the generation of specific questions. Furthermore, we incorporate a data augmentation method to enhance the performance of the generated questions. This approach employs large language models with a designed prompt template to generate additional samples that are relevant to the training dataset. We train and evaluate our model on the three datasets designed for question generation. Our experimental results demonstrate a significant performance improvement compared to the original FLAN-T5 models.