基于自编码器的差分隐私推荐算法

Differential Privacy Recommendation Algorithm Based on Autoencoder

随着深度学习技术在推荐系统领域的快速发展,用户隐私保护问题变得愈发突出。针对这一问题,提出了一种新颖的方法,将差分隐私及自编码器技术应用于深度学习模型训练中,并设计了DPAutoRec算法。该算法通过向梯度中添加符合差分隐私条件的噪声,有效实现了用户隐私的保护。在Movielens-1M数据集上进行了广泛的实验验证,并与传统的AutoRec算法进行了对比,结果表明,在确保差分隐私的前提下,DPAutoRec能够提供有价值的预测结果,为用户隐私和个性化推荐的平衡提供了新的思路与解决方案。

With the rapid development of deep learning technology in the field of recommendation systems,the problem of user privacy protection becomes more and more prominent.To address this problem,a novel method is proposed to apply differential privacy and autoencoder technologies to deep learning model training,and the DPAutoRec algorithm is designed.The algorithm effectively protects user privacy by adding noise that meets the differential privacy condition to the gradient.Extensive experimental verification is carried out on the Movielens-1M dataset,and the experimental results are compared with the conventional AutoRec algorithm.The results indicates that DPAutoRec can still provide valuable prediction results under the premise of ensuring differential privacy,which provides a new idea and solution for the balance between user privacy and personalized recommendation.