A Semantic-Level Plagiarism Detection System using Transformer-Based Language Models

– The rapid growth of digital academic resources has increased the challenge of detecting plagiarism, especially in cases involving paraphrased or structurally modified content. Traditional plagiarism detection systems mainly rely on exact word matching, which often fails to identify semantic-level similarities. This project proposes a web-based plagiarism detection system that uses Natural Language Processing (NLP) techniques with TF-IDF vectorization and Cosine Similarity to detect both direct copying and conceptually similar text. The system improves the identification of partial and paraphrased plagiarism by analyzing contextual similarity between documents. To ensure secure and practical academic deployment, the system incorporates cybersecurity features such as JWTbased authentication, secure file upload validation, and access control. Using a custom dataset of reference and submitted documents, the proposed solution provides a lightweight, interpretable, and efficient approach to plagiarism detection suitable for academic environments.