Federated Personal Knowledge Graph Completion with Lightweight Large Language Models for Personalized Recommendations

TL;DR

FedTREK-LM introduces a federated learning framework combining lightweight large language models and evolving personal knowledge graphs to enable scalable, privacy-preserving personalized recommendations with significant performance improvements.

Contribution

The paper presents FedTREK-LM, a novel framework integrating lightweight LLMs, PKGs, and federated learning for personalized recommendations, outperforming existing methods.

Findings

Over 4x F1-score improvement over baselines

Real user data significantly enhances personalization

Synthetic data reduces performance by up to 46%

Abstract

Personalized recommendation increasingly relies on private user data, motivating approaches that can adapt to individuals without centralizing their information. We present Federated Targeted Recommendations with Evolving Knowledge graphs and Language Models (FedTREK-LM), a framework that unifies lightweight large language models (LLMs), evolving personal knowledge graphs (PKGs), federated learning (FL), and Kahneman-Tversky Optimization to enable scalable, decentralized personalization. By prompting LLMs with structured PKGs, FedTREK-LM performs context-aware reasoning for personalized recommendation tasks such as movie and recipe suggestions. Across three lightweight Qwen3 models (0.6B, 1.7B, 4B), FedTREK-LM consistently and substantially outperforms state-of-the-art KG completion and federated recommendation baselines (HAKE, KBGAT, and FedKGRec), achieving more than a 4x improvement in F1-score on the movie and food benchmarks. Our results further show that real user data is critical for effective personalization, as synthetic data degrades performance by up to 46%. Overall, FedTREK-LM offers a practical paradigm for adaptive, LLM-powered personalization that generalizes across decentralized, evolving user PKGs.