FedGR: Federated Graph Neural Network for Recommendation Systems
Social recommendation systems based on the graph neural network (GNN) have received a lot of research-related attention recently because they can use social information to improve recommendation accuracy and because of the benefits derived from the excellent performance of the graph neural network i...
Main Authors: | , , , |
---|---|
Format: | Article |
Language: | English |
Published: |
MDPI AG
2023-02-01
|
Series: | Axioms |
Subjects: | |
Online Access: | https://www.mdpi.com/2075-1680/12/2/170 |
_version_ | 1797622334561976320 |
---|---|
author | Chuang Ma Xin Ren Guangxia Xu Bo He |
author_facet | Chuang Ma Xin Ren Guangxia Xu Bo He |
author_sort | Chuang Ma |
collection | DOAJ |
description | Social recommendation systems based on the graph neural network (GNN) have received a lot of research-related attention recently because they can use social information to improve recommendation accuracy and because of the benefits derived from the excellent performance of the graph neural network in graphic data modeling. A large number of excellent studies in this area have been proposed one after another, but they all share a common requirement that the data should be centrally stored. In recent years, there have been growing concerns about data privacy. At the same time, the introduction of numerous stringent data protection regulations, represented by general data protection regulations (GDPR), has challenged the recommendation models with conventional centralized data storage. For the above reasons, we have designed a flexible model of recommendation algorithms for social scenarios based on federated learning. We call it the federated graph neural network for recommendation systems (FedGR). Previous related work in this area has only considered GNN, social networks, and federated learning separately. Our work is the first to consider all three together, and we have carried out a detailed design for each part. In FedGR, we used the graph attention network to assist in modeling the implicit vector representation learned by users from social relationship graphs and historical item graphs. In order to protect data privacy, we used FedGR flexible data privacy protection by incorporating traditional cryptography encryption techniques with the proposed “noise injection” strategy, which enables FedGR to ensure data privacy while minimizing the loss of recommended performance. We also demonstrate a different learning paradigm for the recommendation model under federation. Our proposed work has been validated on two publicly available popular datasets. According to the experimental results, FedGR has decreased MAE and RMSE compared with previous work, which proves its rationality and effectiveness. |
first_indexed | 2024-03-11T09:09:47Z |
format | Article |
id | doaj.art-3a7bf55a6b0443fdaabd360c9f867b98 |
institution | Directory Open Access Journal |
issn | 2075-1680 |
language | English |
last_indexed | 2024-03-11T09:09:47Z |
publishDate | 2023-02-01 |
publisher | MDPI AG |
record_format | Article |
series | Axioms |
spelling | doaj.art-3a7bf55a6b0443fdaabd360c9f867b982023-11-16T19:06:15ZengMDPI AGAxioms2075-16802023-02-0112217010.3390/axioms12020170FedGR: Federated Graph Neural Network for Recommendation SystemsChuang Ma0Xin Ren1Guangxia Xu2Bo He3School of Software Engineering, Chongqing University of Posts and Telecommunications, Chongqing 400065, ChinaSchool of Software Engineering, Chongqing University of Posts and Telecommunications, Chongqing 400065, ChinaThe Cyberspace Institute of Advanced Technology, Guangzhou University, Guangzhou 510006, ChinaSchool of Software Engineering, Chongqing University of Posts and Telecommunications, Chongqing 400065, ChinaSocial recommendation systems based on the graph neural network (GNN) have received a lot of research-related attention recently because they can use social information to improve recommendation accuracy and because of the benefits derived from the excellent performance of the graph neural network in graphic data modeling. A large number of excellent studies in this area have been proposed one after another, but they all share a common requirement that the data should be centrally stored. In recent years, there have been growing concerns about data privacy. At the same time, the introduction of numerous stringent data protection regulations, represented by general data protection regulations (GDPR), has challenged the recommendation models with conventional centralized data storage. For the above reasons, we have designed a flexible model of recommendation algorithms for social scenarios based on federated learning. We call it the federated graph neural network for recommendation systems (FedGR). Previous related work in this area has only considered GNN, social networks, and federated learning separately. Our work is the first to consider all three together, and we have carried out a detailed design for each part. In FedGR, we used the graph attention network to assist in modeling the implicit vector representation learned by users from social relationship graphs and historical item graphs. In order to protect data privacy, we used FedGR flexible data privacy protection by incorporating traditional cryptography encryption techniques with the proposed “noise injection” strategy, which enables FedGR to ensure data privacy while minimizing the loss of recommended performance. We also demonstrate a different learning paradigm for the recommendation model under federation. Our proposed work has been validated on two publicly available popular datasets. According to the experimental results, FedGR has decreased MAE and RMSE compared with previous work, which proves its rationality and effectiveness.https://www.mdpi.com/2075-1680/12/2/170social recommendationgraph neural networkfederated learningprivacy protection |
spellingShingle | Chuang Ma Xin Ren Guangxia Xu Bo He FedGR: Federated Graph Neural Network for Recommendation Systems Axioms social recommendation graph neural network federated learning privacy protection |
title | FedGR: Federated Graph Neural Network for Recommendation Systems |
title_full | FedGR: Federated Graph Neural Network for Recommendation Systems |
title_fullStr | FedGR: Federated Graph Neural Network for Recommendation Systems |
title_full_unstemmed | FedGR: Federated Graph Neural Network for Recommendation Systems |
title_short | FedGR: Federated Graph Neural Network for Recommendation Systems |
title_sort | fedgr federated graph neural network for recommendation systems |
topic | social recommendation graph neural network federated learning privacy protection |
url | https://www.mdpi.com/2075-1680/12/2/170 |
work_keys_str_mv | AT chuangma fedgrfederatedgraphneuralnetworkforrecommendationsystems AT xinren fedgrfederatedgraphneuralnetworkforrecommendationsystems AT guangxiaxu fedgrfederatedgraphneuralnetworkforrecommendationsystems AT bohe fedgrfederatedgraphneuralnetworkforrecommendationsystems |