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

Federated learning has emerged as a key paradigm in privacy-preserving computing due to its "data usable but not visible" property, enabling users to collaboratively train models without sharing raw data. Motivated by this, federated recommendation systems offer a promising architecture that balances user privacy with recommendation accuracy through distributed collaborative learning. However, existing federated recommendation methods often neglect the underlying semantic or behavioral relationships between users during parameter aggregation, which limits their recommendation effectiveness. To overcome this limitation, graph-based federated recommendation systems have been proposed to leverage neighborhood information. Yet, conventional graph construction methods usually require access to raw user data or explicit social links, which contradicts the strict privacy requirements of federated learning. In this work, we propose UFGraphFR (User Text-feature-based Graph Federated Recommendation), a novel personalized federated recommendation framework that constructs a user graph based on clients' locally embedded text features. Our core assumption is that users with similar textual feature descriptions exhibit similar preferences. Accordingly, UFGraphFR introduces two key components: (1) a privacy-preserving user relationship graph constructed from the joint embedding layer's weight matrix without leaking raw user attributes; (2) a Transformer-based architecture to model temporal dependencies in user-item interaction sequences. Experimental results on benchmark datasets such as MovieLens and HetRec2011 demonstrate that UFGraphFR achieves recommendation accuracy comparable to both centralized and state-of-the-art federated baselines while preserving user privacy. The code is available at: https://github.com/trueWangSyutung/UFGraphFR.