Scholarlink

This paper introduces an AI-powered platform designed to enhance educational feedback in higher education by combining natural language processing (NLP) and large language models (LLMs). Developed through user-centered design, the system addresses scalability and personalization challenges with two key features: (1) semantic similarity-based tag recommendations (using SpaCy and Levenshtein distance) to identify recurring errors, (2) DeepSeek-V3-generated feedback with instructor co-editing capabilities. Several studies were conducted using the developed platform, aiming to evaluate it from both the teacher’s and student’s perspectives regarding the features provided by the tool. Concerning teachers, it was observed that perceived usefulness, intention to use, task relevance, and feedback quality were positively evaluated. Regarding students, it was noted that the provided feedback was positively assessed, and in most cases, students could not distinguish who generated it (teacher or LLM).The results highlight the platform’s potential to streamline assessment workflows while preserving pedagogical personalization. Limitations include a STEM-oriented participant sample and interface refinements needed for broader adoption. Future work will explore learning outcome impacts and interdisciplinary applications.

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

Tutoria: Delivering Personalized Feedback at Scale with Artificial Intelligence

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Newarney Torrezão da Costa,
  • Cleon Xavier,
  • Fabíola Gonçalves Coelho Ribeiro,
  • Gabriel Alves de Albuquerque Júnior,
  • Luiz Rodrigues,
  • Taciana Pontual da Rocha Falcão,
  • Rafael Ferreira Mello

摘要

This paper introduces an AI-powered platform designed to enhance educational feedback in higher education by combining natural language processing (NLP) and large language models (LLMs). Developed through user-centered design, the system addresses scalability and personalization challenges with two key features: (1) semantic similarity-based tag recommendations (using SpaCy and Levenshtein distance) to identify recurring errors, (2) DeepSeek-V3-generated feedback with instructor co-editing capabilities. Several studies were conducted using the developed platform, aiming to evaluate it from both the teacher’s and student’s perspectives regarding the features provided by the tool. Concerning teachers, it was observed that perceived usefulness, intention to use, task relevance, and feedback quality were positively evaluated. Regarding students, it was noted that the provided feedback was positively assessed, and in most cases, students could not distinguish who generated it (teacher or LLM).The results highlight the platform’s potential to streamline assessment workflows while preserving pedagogical personalization. Limitations include a STEM-oriented participant sample and interface refinements needed for broader adoption. Future work will explore learning outcome impacts and interdisciplinary applications.