| 摘要: |
| 链路预测旨在挖掘或预测隐藏的或即将出现的链路或链接，已被广泛应用到诸多实际网络系统，为用户挖掘潜在的高价值的关联关系。然而，链路预测也可被攻击者用来攻击隐藏的敏感链接，泄露用户隐私，给用户与社会带来难以估量的损失。因此，近年来，链路预测的安全性引起了广大科研工作者的关注，我们称之为逆链路预测问题。逆链路预测的核心思想为通过扰动一定数量的链路来降低链路预测方法对已隐藏的敏感链路的预测概率，从而实现一定的安全性目标。现有机制从研究视角不同可主要分为三类：基于对抗的逆链路预测模型，基于鲁棒性攻击的逆链路预测分析，基于隐私保护的逆链路预测研究。本论文系统综述目前链路预测、逆链路预测、逆链路预测防御方法的研究进展，并对未来研究热点进行了展望，为未来链路预测安全性研究提供基础。 |
| 关键词: 逆链路预测 鲁棒性 链路预测对抗 隐私保护 |
| DOI：10.19363/J.cnki.cn10-1380/tn.2021.03.03 |
| 投稿时间：2020-04-07修订日期：2020-05-20 |
| 基金项目:本课题得到国家自然科学基金（No.61502375）与陕西省自然科学基础研究计划（No.2020JM-203）资助。 |
|
| A Survey of Reverse Link Prediction Methods on Graphs |
| LI Jing,JIANG Zhongyuan,MA Jianfeng |
| School of Cyber Engineering, Xidian University, Xi'an 710071, China |
| Abstract: |
| Link prediction aims to mine or predict links that have been hidden or will emerge in the near future. It has been widely used in many real network systems to mine potential high-value associations for users. However, link prediction can also be used by attackers to attack hidden sensitive links, which lead to the disclosure of user privacy and cause incalculable losses to users and society. Therefore, in recent years, the security of link prediction has attracted the attention of a large number of researchers, and we call it the reverse link prediction problem. The core idea of reverse link prediction is to reduce the probability of hidden sensitive links being predicted by the link prediction methods, which can be achieved by perturbing a certain number of links, and finally achieve the goal of security. The existing mechanisms can be divided into three categories from different research perspectives:the reverse link prediction models based on adversarial attack, the reverse link prediction analysis based on robustness and the reverse link prediction research based on privacy protection. This paper reviews the current research progress of link prediction, reverse link prediction and the defense of reverse link prediction methods. We also discuss the possible future research directions and provides a basis for future link prediction security research. |
| Key words: reverse link prediction robustness adversarial link prediction privacy protection |
