【摘要】 随着数据脱敏技术的持续进步，精确识别隐私数据已成为关键挑战.目前，隐私信息抽取算法主要基于传统自然语言处理技术，如双向循环神经网络和基于注意力机制的预训练语言模型(如BERT).这些模型利用其强大的上下文特征表示能力，克服了传统方法在多义词表示方面的限制.然而，它们在精确判断实体边界方面仍有改进空间.提出了一种新颖的隐私信息抽取算法，该算法融合结构先验知识，通过一种隐私数据结构知识增强机制，提高模型对句子语义结构的理解，从而提高了隐私信息边界判断的准确性.此外，还在多个公开数据集上对模型进行评估，详细的实验结果展示了其有效性.更多还原

【Abstract】 With the continuous advancement of data anonymization technology, accurately identifying private data has become a key challenge. Currently, privacy information extraction algorithms are primarily based on traditional natural language processing techniques, such as bidirectional recurrent neural networks and attention mechanism-based pretrained language models(like BERT and its variants). These models leverage their powerful ability to represent contextual features, overcoming the limitations of traditional methods in representing polysemous words. However, there is still room for improvement in their ability to accurately determine entity boundaries. This study proposes a novel privacy information extraction algorithm that integrates structural prior knowledge and a unique privacy data structural knowledge enhancement mechanism, enhancing the model’s understanding of sentence semantic structures, thereby improving the accuracy of privacy information boundary determination. Moreover, we have evaluated the model on multiple public datasets and provided a detailed analysis of the experimental results, demonstrating its effectiveness.更多还原