EduGuard-LLM: An AI-Generated Content Detector Using Large Language Models for Safeguarding Educational Integrity

In response to the widespread use of AI-generated tools by students to complete assignments, which poses significant challenges to educational integrity and fairness, this study proposes a novel detection model called EduGuard-LLM. EduGuard-LLM leverages the powerful text recognition capabilities of large language models to accurately distinguish between student-authored content and AI-generated content across different educational stages. By deeply analyzing text content and identifying AI-generated features, this model can effectively detect the authenticity of student submissions in primary, middle, high school, and university levels. In our experiments, we utilized 21 publicly available datasets, comprising a total of 164,543 text samples, covering various types of texts from elementary to university levels. The model achieved an accuracy of 93.96% on the pre-training dataset, and accuracies of 95.01%, 94.64%, 93.97%, and 94.94% on four external validation sets, respectively. The experimental results demonstrate that EduGuard-LLM has high detection accuracy across different educational stages, effectively ensuring the authenticity of student submissions. This provides strong technical support for educational institutions, maintaining educational integrity.