Research and Development of Plagiarism Detection Methods Based on Deep Neural Networks

Abstract

This paper presents a comprehensive review of plagiarism forms in academia, including copy-paste, paraphrasing, mosaic, and idea plagiarism. It analyzes challenges posed by automatic paraphrasing tools, cross-language plagiarism, and multi-source plagiarism. The limitations of traditional detection methods such as string matching, n-gram, and TF-IDF are evaluated. Modern approaches based on deep learning (RNN, LSTM, BERT) and stylometry are discussed, leading to the formulation of a plagiarism detection problem that integrates semantic understanding and stylistic analysis.