The rapid growth of generative artificial intelligence has introduced new challenges in identifying whether textual content is written by humans or generated by machines. This raises serious concerns regarding originality and academic integrity. In this work, we present an AI-driven plagiarism detection framework capable of classifying text into three distinct categories: human-written, AI-generated, and hybrid content. The proposed system focuses on analyzing intrinsic linguistic patterns rather than relying on traditional similarity-based methods. A transformer-based model, GPT-2, is utilized to compute perplexity scores, which reflect the predictability of text sequences. In addition, several statistical and stylistic features, including burstiness, repetition rate, and lexical diversity, are extracted to capture writing behavior. These features are then used to train a Random Forest classifier for multi-class classification. Experimental results demonstrate that the proposed approach achieves strong classification performance and effectively distinguishes between different types of content, including partially AI-generated text. By combining deep learning-based language modeling with conventional machine learning techniques, the system improves detection reliability and scalability. Overall, the proposed solution offers a practical approach for identifying AI-assisted plagiarism and supports institutions in maintaining content authenticity in modern digital environments.
Latha et al. (Thu,) studied this question.