基于云联邦的差分隐私保护动态推荐模型

doi:10.3969/j.issn.1671-1122.2023.07.004

摘要/Abstract

摘要：

为提高推荐系统的准确性和个性化水平，同时保护用户的隐私，文章提出一种基于云联邦的差分隐私保护动态推荐模型（P2RCF）。该模型采用注意力机制动态调整融合长短期用户兴趣，增强推荐系统的灵活性，同时引入差分隐私技术和云联邦技术保护用户的隐私信息。文章在公共数据集上进行了实验，实验结果表明，该模型可以在保护用户数据隐私的同时提高推荐的准确性和个性化水平。

关键词: 推荐系统, 云联邦, 注意力机制, 差分隐私, 长短期兴趣

Abstract:

This paper proposed a cloud-based federated differential privacy-Preserving dynamic recommendation model (P2RCF). The model employed an attention mechanism to dynamically adjust the fusion of short-term and long-term user interests, increasing the flexibility of the recommendation system. The paper also introduced differential privacy and cloud federation technologies to protect user privacy information. Experimental evaluations were conducted on public datasets to assess the performance of the proposed model. The results demonstrate that the model improves recommendation accuracy and personalization while preserving user data privacy.

Key words: recommendation system, cloud federation, attention mechanism, differential privacy, long-term and short-term interest

中图分类号:

TP309

刘刚, 杨雯莉, 王同礼, 李阳. 基于云联邦的差分隐私保护动态推荐模型[J]. 信息网络安全, 2023, 23(7): 31-43.

LIU Gang, YANG Wenli, WANG Tongli, LI Yang. Differential Privacy-Preserving Dynamic Recommendation Model Based on Cloud Federation[J]. Netinfo Security, 2023, 23(7): 31-43.

图/表 13

表1

图1

图2

表2

表3

图3

图4

表4

图5

图6

图7

表5

表6

参考文献 25

[1]	SUN Ke, QIAN Tieyun, CHEN Tong, et al. Where to Go Next: Modeling Long-and Short-Term User Preferences for Point-of-Interest Recommendation[C]// AAAI. AAAI Conference on Artificial Intelligence. New York: AAAI, 2020: 214-221.
[2]	CHEN Ming, WEI Zhewei, HUANG Zengfeng, et al. Simple and Deep Graph Convolutional Networks[C]// IMLS. 37th International Conference on Machine Learning. Vienna: PMLR, 2020: 1725-1735.
[3]	WU Libing, QUAN Cong, LI Chenliang, et al. A Context-Aware User-Item Representation Learning for Item Recommendation[J]. ACM Transactions on Information Systems, 2019, 37(2): 1-29.
[4]	CHEN Xusong. Research on Recommendation Algorithm Based on User Behavior Sequence Modeling[D]. Hefei: University of Science and Technology of China, 2021.
	陈旭松. 基于用户行为序列建模的推荐算法研究[D]. 合肥: 中国科学技术大学, 2021.
[5]	ZHANG Liyan. Research on Recommendation Algorithm Based on Sequence Feature Extraction[D]. Chengdu: University of Electronic Science and Technology, 2021.
	张李燕. 基于序列特征提取的推荐算法研究[D]. 成都: 电子科技大学, 2021.
[6]	LIN Guanyu, LIANG Feng, PAN Weike, et al. Fedrec: Federated Recommendation with Explicit Feedback[J]. IEEE Intelligent Systems, 2020, 36(5): 21-30. doi: 10.1109/MIS.2020.3017205 URL
[7]	LIANG Feng, PAN Weike, MING Zhong. Fedrec++: Lossless Federated Recommendation with Explicit Feedback[C]// AAAI. The AAAI Conference on Artificial Intelligence. New York: AAAI, 2021: 4224-4231.
[8]	KHAN F K, FLANAGAN A, Tan K E, et al. A Payload Optimization Method for Federated Recommender Systems[C]// ACM. 15th ACM Conference on Recommender Systems. New York: ACM, 2021: 432-442.
[9]	LUO Sichun, XIAO Yuanzhang, SONG Linqi. Personalized Federated Recommendation via Joint Representation Learning, User Clustering, and Model Adaptation[C]// ACM. 31st ACM International Conference on Information& Knowledge Management. New York: ACM, 2022: 4289-4293.
[10]	SUN Zehua, XU Yonghui, LIU Yong, et al. A Survey on Federated Recommendation Systems[EB/OL]. (2023-03-09) [2023-03-15]. https://arxiv.org/abs/2301.00767.
[11]	LI Xiaoli, LI Yuzheng, WANG Jining, et al. Decentralized Federated Meta-Learning Framework for Few-Shot Multitask Learning[J]. International Journal of Intelligent Systems, 2022, 37(11): 8490-8522. doi: 10.1002/int.v37.11 URL
[12]	ZHANG Feng, XUE Erkang, GUO Ruixin, et al. DS-ADMM++: A Novel Distributed Quantized Admm to Speed up Differentially Private Matrix Factorization[J]. IEEE Transactions on Parallel and Distributed Systems, 2022, 33(6): 1289-1302. doi: 10.1109/TPDS.2021.3110104 URL
[13]	CHEN Chaochao, WU Huiwen, SU Jiajie, et al. Differential Private Knowledge Transfer for Privacy-Preserving Cross-Domain Recommendation[C]// ACM. The ACM Web Conference 2022. New York: ACM, 2022: 1455-1465.
[14]	DENG Wei, CHEN Xiuting, ZHANG Qinghua. Differential Privacy Protection Algorithm Based on Tree Model[J]. Journal of Chongqing University of Posts and Telecommunications, 2020, 32(5): 848-856.
	邓蔚, 陈秀婷, 张清华. 基于树模型的差分隐私保护算法[J]. 重庆邮电大学学报, 2020, 32(5):848-856.
[15]	GUO Taolin, LUO Junzhou, DONG Kai, et al. Locally Differentially Private Item-Based Collaborative Filtering[J]. Information Science. 2019, 502: 229-246. doi: 10.1016/j.ins.2019.06.021 URL
[16]	CHEN Gao, CHAO Huang, LIN Dongsheng, et al. DPLCF Differentially Private Local Collaborative Filtering[C]// ACM. 43rd International ACM SIGIR Conference on Research and Development in Information Retrieval. New York: ACM, 2020: 961-970.
[17]	WANG Yong, YIN Enming, RAN Xun. Differential Privacy Protection Recommendation Algorithm Based on BC Clustering[J]. Journal of Beijing University of Posts and Telecommunications, 2021, 44(2): 81-88. doi: 10.13190/j.jbupt.2020-174
	王永, 尹恩明, 冉旬. 基于BC聚类的差分隐私保护推荐算法[J]. 北京邮电大学学报, 2021, 44(2):81-88. doi: 10.13190/j.jbupt.2020-174
[18]	ZHOU Guorui, ZHU Xiaoqiang, SONG Chengru, et al. Deep Interest Network for Click-Through Rate Prediction[C]// ACM. 24th ACM SIGKDD International Conference on Knowledge Discovery& Data Mining. New York: ACM, 2018: 1059-1068.
[19]	ZHU Yu, LI Hao, LIAO Yikang, et al. What to Do Next: Modeling User Behaviors by Time-LSTM[C]// IJCAI. 26th International Joint Conference on Artificial Intelligence. New York: AAAI, 2017: 3602-3608.
[20]	ZHOU Guorui, MOU Na, FAN Ying, et al. Deep Interest Evolution Network for Click-Through Rate Prediction[C]// AAAI. 33th AAAI Conference on Artificial Intelligence. New York: AAAI, 2019: 5941-5948.
[21]	MOU Luntian, ZHAO Pengfei, XIE Haitao, et al. T-LSTM: A Long Short-Term Memory Neural Network Enhanced by Temporal Information for Traffic Flow Prediction[J]. IEEE Access, 2019, 7: 98053-98060. doi: 10.1109/ACCESS.2019.2929692
[22]	CAO Yi, ZHANG Weifeng, SONG Bo, et al. Position Aware Context Attention for Session-Based Recommendation[J]. Neurocomputing, 2020, 376: 65-72. doi: 10.1016/j.neucom.2019.09.016 URL
[23]	ZHANG Jianqing, WANG Dongjing, YU Dongjin. TLSAN: Time-Aware Long-and Short-Term Attention Network for Next-Item Recommendation[J]. Neurocomputing, 2021, 44(2): 179-191.
[24]	TANG Jiakai, JIN Jiahui, MIAO Zijia, et al. Region-Aware POI Recommendation with Semantic Spatial Graph[C]// IEEE. 24th International Conference on Computer Supported Cooperative Work in Design. New York: IEEE, 2021: 214-219.
[25]	LI Xiang, YANG Xingyao, YU Jiong. Double End Knowledge Graph Convolutional Networks for Recommender Systems[J]. Journal of Frontiers of Computer Science and Technology, 2022, 16(1): 176-184.
	李想, 杨兴耀, 于炯. 基于知识图谱卷积网络的双端推荐算法[J]. 计算机科学与探索, 2022, 16(1):176-184. doi: 10.3778/j.issn.1673-9418.2103072

符号	说明
$U=({{u}_{1}},{{u}_{2}},\cdots,{{u}_{n}})$	用户集合
$I=({{i}_{1}},{{i}_{2}},\cdots,{{i}_{m}})$	物品集合
n	用户集合总数
m	物品集合总数
$B(u)=\{(i_{1}^{u},t_{1}^{u}),(i_{2}^{u},t_{2}^{u}),\cdots,(i_{\|B(u)\|}^{u},t_{\|B(u)\|}^{u})\}$	用户的历史交互行为
$B(u)$	用户的历史交互行为序列总数
$\tau $	时间间隔
p	序列长度
$p_{u}^{long}$	用户长期兴趣备选集
$p_{u}^{short}$	用户短期兴趣备选集

数据集	用户总数/个	商品总数/个	类目总数/个	交互实例数/个
Electronics	192000	63000	704000	2993000
Movies and TV	123000	50000	163000	3147000
CDs and Vinyl	75000	64000	343000	2044000
Subset	100000	1200000	12024000	6716000

模型	评估指标	Electronics	Movies and TV	CDs and Vinyl	Subset
ASVD	AUC	0.7727	0.8156	0.8863	0.8060
DIN		0.7927	0.8388	0.9111	0.8293
DIEN		0.7209	0.8438	0.9128	0.8361
T-LSTM		0.8212	0.8660	0.9181	0.8387
PACA		0.8913	0.8711	0.8832	0.8217
TLSAN		0.9230	0.8986	0.9651	0.8451
RA-NGCF		0.9254	0.8876	0.9400	0.8421
DEKGCN		0.9014	0.8461	0.9036	0.8374
P2RCF		0.9257	0.8751	0.9139	0.8451
ASVD	F1	0.7255	0.7539	0.8128	0.7427
DIN		0.7349	0.7660	0.8348	0.7599
DIEN		0.7327	0.7755	0.8374	0.7632
T-LSTM		0.7448	0.7742	0.8352	0.7591
PACA		0.7434	0.7549	0.8492	0.7497
TLSAN		0.7812	0.7820	0.8716	0.7737
RA-NGCF		0.7816	0.7613	0.8543	0.7784
DEKGCN		0.7423	0.7571	0.8293	0.7695
P2RCF		0.7858	0.7820	0.8329	0.7785

数据集	评估指标	P2RC	P2RCF	diff
Electronics	AUC	0.9296	0.9257	0.4195%
Electronics	F1	0.7886	0.7858	0.3551%
Movies and TV	AUC	0.8769	0.8751	0.2053%
Movies and TV	F1	0.7823	0.7820	0.0383%
CDs and Vinyl	AUC	0.9191	0.9139	0.5658%
CDs and Vinyl	F1	0.8406	0.8329	0.9160%
Subset	AUC	0.8453	0.8451	0.0237%
Subset	F1	0.7787	0.7785	0.0257%

模型	Electronics	Movies and TV	CDs and Vinyl	Subset
LSTM	0.7859	0.8414	0.9086	0.8272
LSTM-T	0.8212	0.8660	0.9181	0.8387
P2RCF-T	0.8248	0.8692	0.9190	0.8402
P2RCF	0.9257	0.8751	0.9139	0.8451