摘要
大多数现有的隐私保护推荐算法是针对显式反馈行为数据的单一推荐算法,且仅适用于可信服务器场景.针对以上不足,提出了一个新的隐私保护推荐系统框架.框架利用用户隐式反馈行为数据,在客户端考虑数值敏感度和数据分布不同,使用LCF-VDP(local collaborative filtering-value differential privacy)机制扰动原始数据并上传到服务器;服务器混合两种算法的相似度,最终选择topk混合相似度发送给每个用户设备,在每个用户设备中进行预测评分计算并推荐.仿真结果表明,提出的方法可以根据不同的需求来选择合适的参数,以达到最佳推荐效果,且LCF-VDP在各种隐私预算下比传统的扰动机制效用更好.
Abstract
Most existing privacy-preserving recommendation algorithms are designed for explicit feedback behavior data and are only applicable in trusted server scenarios.To address these limitations,a new privacy-preser-ving recommendation system framework has been proposed.This framevork utilizes user·s implicit feedback behavior data and considers the different numerical sensitivity and data distributions on the client side.It emplays the local col-laborative filtering-value differential privacy(LCF-VDP)mechanism to perturb the original data and uploading it to the server.The similarity of the two algorithms is mixed at the server,and topk mixed similarity is finally selected and sent to each user device.The prediction score is calculated and recommended in each user device.Experimental re-sults show that the proposed method can select appropriate parameters according to different requirements to achieve the best recommendation effect,and LCF-VDP has better utility than the traditional perturbation mechanism under different privacy budgets allocation.
基金项目
国家自然科学基金(61962012)
广西自然科学基金(2019GXNSFGA245004)
广西青年创新人才科研专项(AD20297028)
广西壮族自治区青年科学基金(2020GXNSFBA297132)
鹏城实验室重大任务项目(PCL2021A09)
鹏城实验室重大任务项目(PCL2021A02)
鹏城实验室重大任务项目(PCL2022A03)
NETL
NSTL
万方数据