| 引用本文： |
-
郭祯,范晨阳,徐嘉,刘干.基于高效深度神经网络的社交网络隐私数据自动感知模型[J].信息安全学报,2026,11(3):102-114 [点击复制]
- GUO Zhen,FAN Chenyang,XU Jia,LIU Gan.Automatic Sensing Model for Private Data in Social Networks Based on Efficient Deep Neural Networks[J].Journal of Cyber Security,2026,11(3):102-114 [点击复制]
|
|
| 摘要: |
| 目前大多数针对在线社交网络用户的隐私感知算法难以精确识别和定位隐私泄露的具体文本位置,并且在处理嵌套隐私方面表现较差,难以满足用户多样化的隐私保护需求。针对嵌套隐私识别,尽管已有较多研究成果,然而,由于嵌套隐私本身的复杂性,实际应用中依然面临识别准确率低和识别速度慢等挑战。具体而言,嵌套隐私识别不仅仅局限于简单的实体识别任务,还需要深入理解隐私信息的层级关系和上下文。这些层级关系可能包括个人信息、社交关系以及行为模式等,它们在不同情境下的隐私敏感度可能存在显著差异。为了解决这些问题,本文提出了一种基于深度学习的社交网络隐私信息内容提取模型,旨在更精准地识别和定位在线社交网络中隐私泄露的具体文本位置,有效处理嵌套隐私问题,提升隐私识别的准确率与速度。该模型结合了改进的RoformerBERT模型、BI-LSTM模型和高效的全局指针算法(Efficient Global Pointer,EGP)构建了一个基于深度学习的快速隐私实体识别模型(Deep Learning-based Fast Privacy Entity Recognition Model,FPERM)。FPERM模型可以自动感知社交网络中用户共享的隐私信息,并准确定位文本中泄露的敏感信息位置,提升训练速度的同时保持模型性能。实验结果表明,FPERM模型高达94.01%的出色总体准确率,并将提出的模型与当前最先进的方法进行了比较分析,结果表明提出的模型具有更优越的性能以及更强的泛化能力,能够有效保护社交网络的隐私信息。 |
| 关键词: 社交网络 深度学习 隐私感知 嵌套实体 |
| DOI：10.19363/J.cnki.cn10-1380/tn.2026.05.08 |
| 投稿时间：2024-10-13修订日期：2025-02-26 |
| 基金项目:本课题得到国家自然科学基金项目(No.62262013)资助。 |
|
| Automatic Sensing Model for Private Data in Social Networks Based on Efficient Deep Neural Networks |
|
GUO Zhen, FAN Chenyang, XU Jia, LIU Gan
|
| (School of Cyberspace Security, Hainan University, Haikou 570228, China) |
| Abstract: |
| Most of the current privacy-aware algorithms for online social network users have difficulty in accurately identifying and locating the specific text location of privacy leakage, and perform poorly in dealing with nested privacy, making it difficult to meet the diverse privacy protection needs of users. Although there have been many research results on nested privacy recognition, due to the complexity of nested privacy, it still faces challenges such as low recognition accuracy and slow recognition speed in practical applications. Specifically, nested privacy recognition is not only limited to the simple entity recognition task, but also requires an in-depth understanding of the hierarchical relationships and contexts of privacy information. These hierarchical relationships may include personal information, social relationships, and behavioral patterns, etc., and their privacy sensitivities may differ significantly in different contexts. To solve these problems, this paper proposes a deep learning-based content extraction model for social network privacy information, which aims to more accurately identify and locate the specific text location of privacy leakage in online social networks, effectively deal with the nested privacy problem, and improve the accuracy and speed of privacy recognition. The model combines the improved RoformerBERT model, BI-LSTM model, and efficient global pointer algorithm (EGP) to construct a Deep Learning-based Fast Privacy Entity Recognition Model (FPERM).The FPERM model can automatically sense the privacy information shared by users in the social network and accurately locate the location of leaked sensitive information in the text, improving the training speed of the model while maintaining the model performance. The experimental results show that the FPERM model achieves 94.01% in overall accuracy, and the proposed model is compared and analyzed with the current state-of-the-art methods, and the results show that the proposed model has a superior performance as well as a stronger generalization ability, which can effectively protect the private information in social networks. |
| Key words: social networks deep learning privacy-awareness nested entities |
