《计算机应用》唯一官方网站 ›› 2023, Vol. 43 ›› Issue (11): 3346-3350.DOI: 10.11772/j.issn.1001-9081.2023010106

• 2022年全国开放式分布与并行计算学术年会(DPCS 2022) • 上一篇    下一篇

基于差分隐私的广告推荐算法

田蕾1,2, 葛丽娜2,3,4()   

  1. 1.广西民族大学 电子信息学院,南宁 530006
    2.广西民族大学 网络通信工程重点实验室,南宁 530006
    3.广西民族大学 人工智能学院,南宁 530006
    4.广西混杂计算与集成电路设计分析重点实验室(广西民族大学),南宁 530006
  • 收稿日期:2023-02-10 修回日期:2023-04-10 接受日期:2023-04-11 发布日期:2023-11-14 出版日期:2023-11-10
  • 通讯作者: 葛丽娜
  • 作者简介:田蕾(1998—),女,山东邹城人,硕士研究生,CCF会员,主要研究方向:差分隐私、推荐算法
    葛丽娜(1969—),女,广西环江人,教授,博士,CCF高级会员,主要研究方向:信息安全、机器学习。 66436539@qq.com
  • 基金资助:
    国家自然科学基金资助项目(61862007)

Advertising recommendation algorithm based on differential privacy

Lei TIAN1,2, Lina GE2,3,4()   

  1. 1.College of Electronic Information,Guangxi Minzu University,Nanning Guangxi 530006,China
    2.Key Laboratory of Network Communication Engineering,Guangxi Minzu University,Nanning Guangxi 530006,China
    3.School of Artificial Intelligence,Guangxi Minzu University,Nanning Guangxi 530006,China
    4.Guangxi Key Laboratory of Hybrid Computation and IC Design Analysis (Guangxi Minzu University),Nanning Guangxi 530006,China
  • Received:2023-02-10 Revised:2023-04-10 Accepted:2023-04-11 Online:2023-11-14 Published:2023-11-10
  • Contact: Lina GE
  • About author:TIAN Lei, born in 1998, M. S. candidate. Her research interests include differential privacy, recommendation algorithm.
    GE Lina, born in 1969, Ph. D., professor. Her research interests include information security, machine learning.
  • Supported by:
    National Natural Science Foundation of China(61862007)

摘要:

随着移动互联网行业进入快速发展阶段,用户数据以及浏览数据大幅增加,所以准确把握用户潜在需求和提高广告推荐效果显得极其重要。DeepFM模型作为目前较为先进的推荐方法,可以从原始特征中抽取到各种复杂度特征,但模型没有对数据进行防护。为了在DeepFM模型中实现隐私保护,提出一种基于差分隐私的DeepFM模型——DP-DeepFM,在模型训练过程中将高斯噪声加入Adam优化算法中,并进行梯度裁剪,防止加入噪声过大引发模型性能下降。在广告Criteo数据集上的实验结果表明,与DeepFM相比,DP-DeepFM的准确率仅下降了0.44个百分点,但它能提供差分隐私保护,更具安全性。

关键词: 差分隐私, 推荐算法, 梯度下降, 深度学习, Adam优化算法

Abstract:

With the rapid development of the mobile Internet industry, user data and browsing data have increased significantly, so it is extremely important to accurately grasp the potential needs of users and improve the effect of advertisement recommendation. As a relatively advanced recommendation method at present, DeepFM model can extract various complexity features from the original features, but the model does not protect the data. In order to realize the privacy protection in DeepFM model, a new DeepFM model based on Differential Privacy (DP) was proposed, namely DP-DeepFM. The Gaussian noise was added to Adam optimization algorithm in the training process of DP-DeepFM and the gradient clipping was performed to prevent the addition of excessive noise causing poor model performance. Experimental results on advertising dataset Criteo show that compared with DeepFM, DP-DeepFM only has the accuracy decreased by 0.44 percentage points, but it provides differential privacy protection and is more secure.

Key words: Differential Privacy (DP), recommendation algorithm, gradient descent, deep learning, Adam optimization algorithm

中图分类号: