UFGraphFR: Graph Federation Recommendation System based on User Text description features

TL;DR

UFGraphFR tackles privacy-aware federated recommendations by transforming private user attributes into text descriptions, which are embedded into a semantic space to construct a hidden user-relationship graph on the server. The framework combines a joint embedding layer and Transformer-based local modeling with a graph-convolution-based server aggregation to learn global item embeddings, while preserving data locality and enabling personalized sequence modeling. Key contributions include (1) semantic-text based user graph construction from private data, (2) graph-guided server learning of global item embeddings, (3) Transformer-enhanced sequence modeling on the client, and (4) a local-DP option to strengthen privacy guarantees. Experiments on four benchmarks show UFGraphFR outperforms state-of-the-art centralized and federated baselines, with robust performance across PLMs and a favorable efficiency-accuracy trade-off via UFGraphFR-Lite. The work has practical implications for privacy-preserving, semantically informed federated recommendations and provides a publicly available implementation.

Abstract

Federated learning offers a privacy-preserving framework for recommendation systems by enabling local data processing; however, data localization introduces substantial obstacles. Traditional federated recommendation approaches treat each user as an isolated entity, failing to construct global user relationship graphs that capture collaborative signals, which limits the accuracy of recommendations. To address this limitation, we derive insight from the insight that semantic similarity reflects preference. similarity, which can be used to improve the construction of user relationship graphs. This paper proposes UFGraphFR, a novel framework with three key components: 1) On the client side, private structured data is first transformed into text descriptions. These descriptions are then encoded into semantic vectors using pre-trained models; 2) On the server side, user relationship graphs are securely reconstructed using aggregated model weights without accessing raw data, followed by information propagation through lightweight graph neural networks; 3) On the client side, user behavior sequences are personalized using Transformer architectures. Extensive experiments conducted on four benchmark datasets demonstrate that UFGraphFR significantly outperforms state-of-the-art baselines in both recommendation accuracy and personalization. The framework also maintains robustness across different pre-trained models, as evidenced by the consistent performance metrics obtained. This work provides a practical method for efficient federated recommendations with strict privacy by using semantic vectors, secure user relationship graphs, and personalized behavior sequences. The code is available at: https://github.com/trueWangSyutung/UFGraphFR

Figures (6)

• Figure 1: a) shows the traditional centralized recommendation system. The user client first upload its own user characteristics and interactive data (with or without the consent of the user) to the server, and the server trains a recommendation model in the cloud to recommend suitable items for the user. b) shows a federated learning framework. Each client has a local model. The cloud sends the model parameters to the client in advance, and the client conducts local training without uploading private data to the cloud and uploals the trained model parameters to the server for aggregation.
• Figure 2: This figure shows a recommendation system framework that integrates federated learning and graph neural networks. Local training of users and item embeddings on the client side, uploading model parameters to the server; The server constructs a relationship graph based on user similarity, aggregates and generates global item embeddings, and sends them back to the client to achieve collaborative recommendation under privacy protection.
• Figure 3: Client Model:The image shows our client model, with the yellow box on the left indicating the structure of the joint embedding layer (consisting of the pre-trained language model layer, and linear layer). On the right is the Transformer Block used for sequence modeling.
• Figure 4: Performance comparison on four datasets.
• Figure 5: Ablation study results of UFGraphFR on four datasets. GPFedRec is used as the baseline. Performance drops relative to the full model are shown below each variant.