Style Change-Oriented Text Plagiarism Detection Method
楚淋 罗 · Modeling and Simulation · 2025
近年来，随着大语言模型(LLM)的飞速发展，抄袭检测任务面临着前所未有的挑战。针对这一问题，文章提出了一种面向风格转变的检测模型。所提出模型通过结合BERT与图注意力网络，能够有效学习文本的风格特征并实现风格分类。同时，还巧妙地引入对比学习机制，进一步增强了文本的风格特征表示能力，从而显著提升了模型对写作风格改变的检测性能。实验结果表明，在PAN 2022写作风格改变检测数据集上，本文提出的模型相较于现有代表性方法取得了更优秀的检测效果。此外，通过消融实验验证了风格增强机制的有效性，并证明了图注意力网络在捕捉文本写作风格特征方面的优势。本文提出的方法不仅提高了风格转变检测的准确性，还为后续抄袭检测任务提供了前置条件。In recent years, with the rapid development of large language models (LLM), plagiarism detection is facing unprecedented challenges. To solve this problem, this paper proposes a style change oriented detection model. By combining BERT and graph attention networks, the proposed model can effectively learn text style features and realize style classification. At the same time, it also cleverly introduces a contrastive learning mechanism to further enhance the representation ability of text style features, thus significantly improving the model’s detection performance of writing style changes. The experimental results show that the model proposed in this paper achieves better detection results than the existing representative methods on the PAN 2022 writing style change detection dataset. In addition, the effectiveness of the style enhancement mechanism was verified through ablation experiments, and the advantage of the graph attention network in capturing stylistic features of text writing was demonstrated. The method proposed in this paper not only improves the accuracy of style change detection, but also provides preconditions for subsequent plagiarism detection tasks.