Scholarlink

Systems of education and assessment are mostly based on questions and answers all over the world. They are used to assess and confirm a student’s comprehension of academic subject. In this study we have developed a system that generates Subjective questions based on user query and the document uploaded by the user. The proposed models use a custom dataset, the user must upload their file to obtain the desired result. We started this procedure with a dataset that contained a variety of file types, including PDFs, text files (TXT), CSVs, and few others. Three of the specific gpt4all models—Falcon, Llama-2-7b, and Mini Orca (Large)—was used in this investigation. We get a high-level summary of these models’ capabilities from the comparison of the models. The comparison of the models was done with the help of a field expert who hand annotated each question generated by the proposed models. When comparing the models, we discovered that Llama-2-7b produces questions that are the most relevant to the topic the user has asked about.

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

Subjective Question Bank Generation Using Large Language Models with Custom Knowledge Base

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Amaan Sayed,
  • Mahendra Kanojia,
  • Subhashish Nabajja

摘要

Systems of education and assessment are mostly based on questions and answers all over the world. They are used to assess and confirm a student’s comprehension of academic subject. In this study we have developed a system that generates Subjective questions based on user query and the document uploaded by the user. The proposed models use a custom dataset, the user must upload their file to obtain the desired result. We started this procedure with a dataset that contained a variety of file types, including PDFs, text files (TXT), CSVs, and few others. Three of the specific gpt4all models—Falcon, Llama-2-7b, and Mini Orca (Large)—was used in this investigation. We get a high-level summary of these models’ capabilities from the comparison of the models. The comparison of the models was done with the help of a field expert who hand annotated each question generated by the proposed models. When comparing the models, we discovered that Llama-2-7b produces questions that are the most relevant to the topic the user has asked about.