IP Library › Granted Patent US 12,572,817

Granted Patent B2

US 12,572,817 · App. 18/310,765 · Granted Mar 10, 2026

Federated recommendation system, device, and method

Inventors: Sichun Luo (Kowloon, HK); Linqi Song (Kowloon, HK)

Assignee: City University of Hong Kong

G06N3/098 G06N3/045

View Patent ↗

Loading inventors, assignments & file history…

Monitor This Case

Get email alerts when status or documents change.

Order Certified Copies

Most orders are placed with the USPTO same day — all within 24 business hours.

Order via The Patent Place →

Pre-filled with this patent's details

Quick Facts

Patent No.: US 12,572,817
App. No.: 18/310,765
Granted: Mar 10, 2026
Kind: B2

External Links

USPTO Patent Center↗ Full Patent PDF↗

Abstract

A federated recommendation system with a server and client devices. The server can group client device users into clusters. The server can further: for each respective cluster, process local model parameters associated with local graph neural networks for at least some client device users in the corresponding cluster, to obtain cluster-level model parameters associated with a cluster-level federated model for the corresponding cluster; and process local model parameters associated with local graph neural networks for at least some client device users in each of two or more of the clusters to obtain global model parameters associated with a global federated model. The server can further provide, to a client device, the cluster-level model parameters associated with the corresponding cluster-level federated model and the global model parameters associated with the global federated model, for facilitating generation or update of a personalized recommendation model for the client device user.

Claims (91)

1 . A computer-implemented method for operating a server of a federated recommendation system, comprising:

(a) grouping a plurality of client device users of the federated recommendation system into two or more clusters;

(b) for each respective one of the clusters: processing local model parameters associated with local graph neural networks for at least some of the client device users in the corresponding cluster, to obtain cluster-level model parameters associated with a cluster-level federated model for the corresponding cluster;

(c) processing local model parameters associated with local graph neural networks for at least some of the client device users in each of two or more of the clusters to obtain global model parameters associated with a global federated model for the federated recommendation system; and

(d) providing, to a client device for a client device user of the federated recommendation system, the cluster-level model parameters associated with the cluster-level federated model for the corresponding cluster of the client device user and the global model parameters associated with the global federated model, for facilitating generation or update of a personalized recommendation model for the client device user.

2 . The computer-implemented method of claim 1 ,

wherein the personalized recommendation model for the client device user is operable to provide a recommendation to the client device user; and

wherein the recommendation comprises a predicted rating or preference associated with a plurality of items.

3 . The computer-implemented method of claim 1 ,

wherein the local model parameters associated with each respective local graph neural network respectively comprise gradients of the corresponding local graph neural network;

wherein the cluster-level model parameters associated with each respective cluster-level federated model respectively comprise gradients of the corresponding cluster-level federated model; and/or

wherein the global model parameters associated with the global federated model comprise gradients of the global federated model.

4 . The computer-implemented method of claim 1 , wherein the grouping in (a) is based on user representations of the plurality of client device users of the federated recommendation system.

5 . The computer-implemented method of claim 1 ,

wherein in (b), for at least one of the clusters, the processing of the local model parameters is based on a weighted sum method; and

wherein the weighted sum method applies one or more weightings to one or more of the local graph neural networks for the at least some of the client device users in the corresponding cluster.

6 . The computer-implemented method of claim 1 , wherein (b) comprises:

for at least one of the clusters: processing local model parameters associated with local graph neural networks for only some of the client device users in the corresponding cluster, to obtain cluster-level model parameters associated with a cluster-level federated model for the corresponding cluster.

7 . The computer-implemented method of claim 6 , wherein the only some of the client device users in the corresponding cluster are selected randomly from all the client device users in the corresponding cluster.

8 . The computer-implemented method of claim 1 , wherein (b) comprises:

for each respective one of the clusters: processing local model parameters associated with local graph neural networks for only some of the client device users in the corresponding cluster, to obtain cluster-level model parameters associated with a cluster-level federated model for the corresponding cluster.

9 . The computer-implemented method of claim 8 , wherein the number of client device users selected for each respective one of the clusters is based on the total number of client device users in the corresponding cluster.

10 . The computer-implemented method of claim 1 , wherein (c) comprises:

processing local model parameters associated with local graph neural networks for at least some of the client device users of the federated recommendation system in each respective one of the clusters to obtain the global model parameters associated with the global federated model for the federated recommendation system.

11 . The computer-implemented method of claim 1 ,

wherein (c) comprises: processing local model parameters associated with local graph neural networks for at least some of the client device users of the federated recommendation system in at least two of the clusters using a weighted sum method, to obtain the global model parameters associated with the global federated model for the federated recommendation system; and

wherein the weighted sum method applies one or more weightings to one or more of the local graph neural networks for the at least some of the client device users of the federated recommendation system in the at least two of the clusters.

12 . A computer-implemented method for operating a client device of a federated recommendation system, comprising:

(a) receiving cluster-level model parameters associated with a cluster-level federated model and global model parameters associated with a global federated model, the cluster-level federated model is for a plurality of client device users of the federated recommendation system that belong to the same cluster as the client device user, and the global federated model is for the federated recommendation system; and

(b) generating or updating a personalized recommendation model for a client device user based on the cluster-level model parameters, the global model parameters, and local model parameters associated with a local graph neural network for the client device user;

wherein the personalized recommendation model is arranged to provide a recommendation to the client device user.

13 . The computer-implemented method of claim 12 , wherein the recommendation comprises a predicted rating or preference associated with a plurality of items.

14 . The computer-implemented method of claim 12 ,

wherein the cluster-level model parameters associated with the cluster-level federated model is obtained based on local model parameters associated with local graph neural networks for at least some of the client device users in the corresponding cluster; and

wherein the global model parameters associated with the global federated model is obtained based on local model parameters associated with local graph neural networks for at least some of the client device users in each of two or more of the clusters.

15 . The computer-implemented method of claim 12 ,

wherein the local model parameters associated with the local graph neural network comprise gradients of the local graph neural network;

wherein the cluster-level model parameters associated with the cluster-level federated model comprise gradients of the cluster-level federated model; and/or

wherein the global model parameters associated with the global federated model comprise gradients of the global federated model.

16 . The computer-implemented method of claim 12 ,

wherein (b) comprises generating or updating the personalized recommendation model based on a weighted sum of: the local model parameters, the cluster-level model parameters, and the global model parameters.

17 . A computer-implemented method for operating a client device of a federated recommendation system, comprising:

(a) processing user representations and item representations using a personalized recommendation model customized for a client device user of the federated recommendation system; and

(b) based on the processing, providing a recommendation to the client device user;

wherein the personalized recommendation model is generated or updated based on:

local model parameters associated with a local graph neural network for the client device user;

cluster-level model parameters associated with a cluster-level federated model, the cluster-level federated model is for a plurality of client device users of the federated recommendation system that belong to the same cluster as the client device user; and

global model parameters associated with a global federated model for the federated recommendation system.

18 . The computer-implemented method of claim 17 , wherein the recommendation comprises a predicted rating or preference associated with a plurality of items.

19 . The computer-implemented method of claim 17 , wherein client device users of the federated recommendation system are grouped into two or more clusters based on user representations of the client device users.

20 . The computer-implemented method of claim 19 ,

21 . The computer-implemented method of claim 17 ,

wherein the local model parameters associated with the local graph neural network comprise gradients of the local graph neural network;

wherein the cluster-level model parameters associated with the cluster-level federated model comprise gradients of the corresponding cluster-level federated model; and/or

wherein the global model parameters associated with the global federated model comprise gradients of the global federated model.

22 . The computer-implemented method of claim 17 , wherein the personalized recommendation model is generated or updated based generated or updated on a weighted sum of the local model parameters, the cluster-level model parameters, and the global model parameters.

23 . The computer-implemented method of claim 17 , further comprising:

processing, using the local graph neural network for the client device user,

(i) raw user representations and raw item representations of the client device user, and

(ii) raw user representations and raw item representations of one or more other client device users,

to obtain the user representations and the item representations.

24 . The computer-implemented method of claim 23 , further comprising:

processing user information of a client device user, item information associated with one or more items, and user-item interaction information associated with the client device user's interaction with the one or more items to obtain the raw user representations and the raw item representations.

25 . The computer-implemented method of claim 24 , wherein the processing to obtain the raw user representations and the raw item representations comprises:

processing the user information and the item information to obtain attribute representations; and

processing the attribute representations and the user-item interaction information to obtain the raw user representations and the raw item representations.

26 . The computer-implemented method of claim 25 , wherein the processing of the user information and the item information comprises:

processing the user information and the item information using a model that comprises at least one linear network/layer and at least one feature crossing network/layer.

27 . The computer-implemented method of claim 25 , wherein the processing of the attribute representations and the user-item interaction information comprises:

processing the attribute representations and the user-item interaction information using an attention mechanism.

28 . A server for a federated recommendation system, comprising:

one or more processors; and

memory storing one or more programs configured to be executed by the one or more processors, the one or more programs including instructions for:

grouping a plurality of client device users of the federated recommendation system into two or more clusters;

for each respective one of the clusters: processing local model parameters associated with local graph neural networks for at least some of the client device users in the corresponding cluster, to obtain cluster-level model parameters associated with a cluster-level federated model for the corresponding cluster;

processing local model parameters associated with local graph neural networks for at least some of the client device users in each of two or more of the clusters to obtain global model parameters associated with a global federated model for the federated recommendation system; and

providing, to a client device for a client device user of the federated recommendation system, the cluster-level model parameters associated with the cluster-level federated model for the corresponding cluster of the client device user and the global model parameters associated with the global federated model, for facilitating generation or update of a personalized recommendation model for the client device user.

29 . A client device for a federated recommendation system, comprising:

one or more processors; and

memory storing one or more programs configured to be executed by the one or more processors, the one or more programs including instructions for:

receiving cluster-level model parameters associated with a cluster-level federated model and global model parameters associated with a global federated model, the cluster-level federated model is for a plurality of client device users of the federated recommendation system that belong to the same cluster as the client device user, and the global federated model is for the federated recommendation system; and

generating or updating a personalized recommendation model for a client device user based on the cluster-level model parameters, the global model parameters, and local model parameters associated with a local graph neural network for the client device user;

wherein the personalized recommendation model is arranged for providing a recommendation to the client device user.

30 . A client device for a federated recommendation system, comprising:

one or more processors; and

memory storing one or more programs configured to be executed by the one or more processors, the one or more programs including instructions for:

processing user representations and item representations using a personalized recommendation model customized for a client device user of the federated recommendation system; and

based on the processing, providing a recommendation to the client device user;

wherein the recommendation comprises a predicted rating or preference associated with a plurality of items.

Assignments (1)

ASSIGNMENT OF ASSIGNOR'S INTEREST Recorded May 2, 2023

From: LUO, SICHUN; SONG, LINQI

To: CITY UNIVERSITY OF HONG KONG

Reel/Frame 063507/0025 →

Continuity (2)

Provisional Application 63355798 · Jun 27, 2022

Related Publication 20230419123A1 · Dec 28, 2023

References Cited (22)

US 20140297456A1 · Sapugay · 2014 [cited by examiner]

US 20230021233A1 · Ammad-Ud-Din · 2023 [cited by examiner]

Waqar Ali, Rajesh Kumar, Zhiyi Deng, Yansong Wang, and Jie Shao. 2021. A federated learning approach for privacy protection in context-aware recommender systems. Comput. J. 64, 7 (2021), 1016-1027. [cited by applicant]

Muhammad Ammad-Ud-Din, Elena Ivannikova, Suleiman A Khan, Were Oyomno, Qiang Fu, Kuan Eeik Tan, and Adrian Flanagan. 2019. Federated collaborative filtering for privacy-preserving personalized recommendation system. arX… [cited by applicant]

Di Chai, Leye Wang, Kai Chen, and Qiang Yang. 2020. Secure federated matrix factorization. IEEE Intelligent Systems (2020). [cited by applicant]

Mukund Deshpande and George Karypis. 2004. Item-based top-n recommendation algorithms. ACM Transactions on Information Systems (TOIS) 22, 1 (2004), 143-177. [cited by applicant]

Yongjie Du, Deyun Zhou, Yu Xie, Jiao Shi, and Maoguo Gong. 2021. Federated matrix factorization for privacy-preserving recommender systems. Applied Soft Computing 111 (2021), 107700. [cited by applicant]

Ali Mamdouh Elkahky, Yang Song, and Xiaodong He. 2015. A multi-view deep learning approach for cross domain user modeling in recommendation systems. In Proceedings of the 24th international conference on world wide web.… [cited by applicant]

Yu Guo, Hongcheng Xie, Yinbin Miao, Cong Wang, and Xiaohua Jia. 2020. Fedcrowd: A federated and privacy-breserving crowdsourcing platform on blockchain. IEEE Transactions on Services Computing (2020). [cited by applicant]

Xiangnan He, Kuan Deng, Xiang Wang, Yan Li, Yongdong Zhang, and Meng Wang. 2020. Lightgcn: Simplifying and powering graph convolution network for recommendation. In Proceedings of the 43rd International ACM SIGIR confer… [cited by applicant]

Zitao Li, Bolin Ding, Ce Zhang, Ninghui Li, and Jingren Zhou. 2021. Federated matrix factorization with privacy guarantee. Proceedings of the VLDB Endowment 15, 4 (2021), 900-913. [cited by applicant]

James MacQueen et al. 1967. Some methods for classification and analysis of multivariate observations. In Proceedings of the fifth Berkeley symposium on mathematical statistics and probability, vol. 1. Oakland, CA, USA,… [cited by applicant]

Brendan McMahan, Eider Moore, Daniel Ramage, Seth Hampson, and Blaise Aguera y Arcas. 2017. Communication-efficient learning of deep networks from decentralized data. In Artificial intelligence and statistics. PMLR, 127… [cited by applicant]

Lorenzo Minto, Moritz Haller, Benjamin Livshits, and Hamed Haddadi. 2021. Stronger Privacy for Federated Collaborative Filtering With Implicit Feedback. In Fifteenth ACM Conference on Recommender Systems. 342-350. [cited by applicant]

Khalil Muhammad, Qinqin Wang, Diarmuid O'Reilly-Morgan, Elias Tragos, Barry Smyth, Neil Hurley, James Geraci, and Aonghus Lawlor. 2020. Fedfast: Going beyond average for faster training of federated recommender systems.… [cited by applicant]

Stefano Savazzi, Monica Nicoli, Mehdi Bennis, Sanaz Kianoush, and Luca Barbieri. 2021. Opportunities of federated learning in connected, cooperative, and automated industrial systems. IEEE Communications Magazine 59, 2 … [cited by applicant]

Ben Tan, Bo Liu, Vincent Zheng, and Qiang Yang. 2020. A federated recommender system for online services. In Fourteenth ACM Conference on Recommender Systems. 579-581. [cited by applicant]

Xiang Wang, Xiangnan He, Meng Wang, Fuli Feng, and Tat-Seng Chua. 2019. Neural graph collaborative filtering. In Proceedings of the 42nd international ACM SIGIR conference on Research and development in Information Retr… [cited by applicant]

Chuhan Wu, Fangzhao Wu, Yang Cao, Yongfeng Huang, and Xing Xie. 2021. FedGNN: Federated graph neural network for privacy-preserving recommendation. arXiv preprint arXiv:2102.04925 (2021). [cited by applicant]

Rex Ying, Ruining He, Kaifeng Chen, Pong Eksombatchai, William L Hamilton, and Jure Leskovec. 2018. Graph convolutional neural networks for web-scale recommender systems. In Proceedings of the 24th ACM SIGKDD Internatio… [cited by applicant]

Chen Zhang, Yu Guo, Hongwei Du, and Xiaohua Jia. 2020. Pfcrowd: Privacy-preserving and federated crowdsourcing framework by using blockchain. In 2020 IEEE/ACM 28th International Symposium on Quality of Service (IWQoS). … [cited by applicant]

Tao Qi, Fangzhao Wu, Chuhan Wu, Yongfeng Huang, and Xing Xie. 2020. Fedrec: Privacy-preserving news recommendation with federated learning. arXiv preprint arXiv:2003.09592 (2020). [cited by applicant]