Scholarlink

<p>Online academic and administrative services in higher education increasingly require students, faculty, and staff to complete structured forms across heterogeneous institutional platforms. These tasks are complicated by variations in field naming, page layout, validation rules, and dynamic interaction logic, which limits the reliability of rule-based form-filling tools. To address these challenges, this study proposes KAFA, a knowledge-augmented form-filling agent for higher education services. KAFA integrates a large language model (DeepSeek), Chrome DevTools Protocol (CDP)-based web structure parsing, a multimodal personal knowledge base, tri-source collaborative retrieval, historical trajectory reuse, and adaptive error correction. Experiments on public recruitment forms, which share profile-oriented structures with many higher-education service forms, show that KAFA improves field matching and end-to-end form completion across different complexity levels. The results demonstrate the potential of combining large language models with structured, vector, and graph-based knowledge for robust form filling in education-oriented web services.</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

Knowledge-augmented form-filling agent for higher education services

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年7月6日

查看全文
  • Haiwei Sang,
  • Mengyun Liu,
  • Jing Tang,
  • Yuxiang Yang,
  • Yuling Chen

摘要

Online academic and administrative services in higher education increasingly require students, faculty, and staff to complete structured forms across heterogeneous institutional platforms. These tasks are complicated by variations in field naming, page layout, validation rules, and dynamic interaction logic, which limits the reliability of rule-based form-filling tools. To address these challenges, this study proposes KAFA, a knowledge-augmented form-filling agent for higher education services. KAFA integrates a large language model (DeepSeek), Chrome DevTools Protocol (CDP)-based web structure parsing, a multimodal personal knowledge base, tri-source collaborative retrieval, historical trajectory reuse, and adaptive error correction. Experiments on public recruitment forms, which share profile-oriented structures with many higher-education service forms, show that KAFA improves field matching and end-to-end form completion across different complexity levels. The results demonstrate the potential of combining large language models with structured, vector, and graph-based knowledge for robust form filling in education-oriented web services.