A Large Language Model-Based Plagiarism Detection System Architecture for Academic Journals in Higher Education

In recent years, academic misconduct in universities and research institutions has posed significant challenges to academic integrity and reputation. However, existing plagiarism detection systems exhibit notable limitations in covering dynamic internet content and detecting implicit plagiarism. In this paper we analyze the shortcomings of current detection methods and propose a plagiarism detection system architecture based on search engines and large language models (LLMs). The system utilizes a Transformer-based architecture, integrating Sentence Transformer for semantic text embedding. By leveraging automated retrieval technology, it captures dynamic internet content in real-time and employs semantic similarity calculations to accurately identify potential plagiarisms, paraphrasing, and cross-lingual duplication. With adaptability and robustness, the system effectively addresses dynamic content coverage and enhances implicit plagiarism detection capabilities. It provides technological support for academic review processes in university academic journals and fosters the development of academic norms among potential authors, contributing to the reinforcement of academic integrity within the higher education system.