Personalized Graph-Based Retrieval for Large Language Models

ACL ARR 2025 May Submission6485 Authors

Abstract: As large language models (LLMs) continue to evolve, their ability to deliver personalized, context-aware responses holds significant promise for enhancing user experiences. However, most existing personalization approaches rely solely on user history, limiting their effectiveness in cold-start and sparse-data scenarios. We introduce Personalized Graph-based Retrieval-Augmented Generation (PGraphRAG), a framework that enhances personalization by leveraging user-centric knowledge graphs. By integrating structured user information into the retrieval process and augmenting prompts with graph-based context, PGraphRAG improves both relevance and generation quality. We also present the Personalized Graph-based Benchmark for Text Generation, designed to evaluate personalized generation in real-world settings where user history is minimal. Experimental results show that PGraphRAG consistently outperforms state-of-the-art methods across diverse tasks, achieving average ROUGE-1 gains of 14.8\% on long-text and 4.6\% on short-text generation—highlighting the unique advantages of graph-based retrieval for personalization.

Paper Type: Long

Research Area: NLP Applications

Research Area Keywords: Graphs, personalization, text generation

Contribution Types: Model analysis & interpretability, NLP engineering experiment, Publicly available software and/or pre-trained models, Data resources, Data analysis

Languages Studied: English

Submission Number: 6485