Using Large Language Models for Students’ Essays Plagiarism Detection

Large language models are used and fine-tuned today for different tasks. In educational environment, one problematic usage of this technology is for creation of plagiarism work. The focus of this work is to use large language models to detect plagiarised student essays. The recent relevant Kaggle competition "Detect AI Generated Text" is analysed. The top most accurate solutions in this competition used the ensembles of various large language models. In the efficiency track of the competition, winning solutions used more classical machine learning methods. We also analyse the main datasets used to finetune the LLM and machine learning models for students’ essays plagiarism detection. The most accurate models use comprehensive set of essay data and ensemble of different large language models.