This paper proposes a user-governed data contribution federated recommendation framework (UGFedRec) to address the limitations of existing federated recommendation systems (FedRecs), which assume all users have the same 0-privacy budget. UGFedRec allows users to control whether and how much data they share with the server, enabling personalized privacy settings. The proposed method, CDCGNNFed, combines cloud-device collaborative graph neural networks to train user-centric ego graphs locally and high-order graphs based on shared data. A graph mending strategy is used to predict missing links in the graph, enhancing the effectiveness of graph neural networks. The model also employs contrastive learning to align local and global views of the same node, improving recommendation performance. Extensive experiments on two public datasets show that the proposed method outperforms existing baselines in most scenarios, particularly in partial uploading cases. The results demonstrate that UGFedRec effectively balances privacy concerns with recommendation utility, offering a more flexible and personalized approach to federated learning in recommendation systems.This paper proposes a user-governed data contribution federated recommendation framework (UGFedRec) to address the limitations of existing federated recommendation systems (FedRecs), which assume all users have the same 0-privacy budget. UGFedRec allows users to control whether and how much data they share with the server, enabling personalized privacy settings. The proposed method, CDCGNNFed, combines cloud-device collaborative graph neural networks to train user-centric ego graphs locally and high-order graphs based on shared data. A graph mending strategy is used to predict missing links in the graph, enhancing the effectiveness of graph neural networks. The model also employs contrastive learning to align local and global views of the same node, improving recommendation performance. Extensive experiments on two public datasets show that the proposed method outperforms existing baselines in most scenarios, particularly in partial uploading cases. The results demonstrate that UGFedRec effectively balances privacy concerns with recommendation utility, offering a more flexible and personalized approach to federated learning in recommendation systems.