Comparing Retrieval-Augmentation and Parameter-Efficient Fine-Tuning for Privacy-Preserving Personalization of Large Language Models

2024-09-14

arXiv.org

AI Summary

This study addresses user privacy preservation in large language model (LLM) personalization, systematically comparing retrieval-augmented generation (RAG) and parameter-efficient fine-tuning (PEFT) under privacy-sensitive settings. Using seven diverse datasets and multi-task benchmarks, we empirically establish their complementary performance boundaries: RAG significantly outperforms PEFT in cold-start and few-shot scenarios (average +14.92%), while PEFT gains effectiveness with increasing user data volume (+1.07% on average); their integration further improves performance to +15.98%. We propose two novel mechanisms—private user-data indexing and controllable personalized prompt construction—that enable efficient, interpretable personalization without data leaving the user domain. Our findings provide both methodological guidance and empirical baselines for privacy-first LLM customization, advancing the design of secure, adaptive, and transparent personalized LLM systems.

Technology Category

• Machine Learning: Privacy
• Data Mining & Knowledge Management: Conversational Systems for Recommendation & Retrieval
• Natural Language Processing: (Large) Language Models

Application Category

• User Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendation
• Search and Retrieval-Augmented AI: Retrieval-Augmented Generation (RAG) and multi-modal RAG
• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions

Abstract

Privacy-preserving methods for personalizing large language models (LLMs) are relatively under-explored. There are two schools of thought on this topic: (1) generating personalized outputs by personalizing the input prompt through retrieval augmentation from the user's personal information (RAG-based methods), and (2) parameter-efficient fine-tuning of LLMs per user that considers efficiency and space limitations (PEFT-based methods). This paper presents the first systematic comparison between two approaches on a wide range of personalization tasks using seven diverse datasets. Our results indicate that RAG-based and PEFT-based personalization methods on average yield 14.92% and 1.07% improvements over the non-personalized LLM, respectively. We find that combining RAG with PEFT elevates these improvements to 15.98%. Additionally, we identify a positive correlation between the amount of user data and PEFT's effectiveness, indicating that RAG is a better choice for cold-start users (i.e., user's with limited personal data).

Problem

• Compares RAG and PEFT for privacy-preserving LLM personalization
• Evaluates effectiveness of combining RAG and PEFT methods
• Analyzes data volume impact on PEFT performance

Innovation

• Uses retrieval-augmented generation for personalization
• Employs parameter-efficient fine-tuning for user adaptation
• Combines RAG and PEFT for enhanced performance