A Survey of Personalization: From RAG to Agent

Abstract: Personalization has become an essential capability in modern AI systems, enabling customized interactions that align with individual user preferences, contexts, and goals. Recent research has increasingly concentrated on Retrieval-Augmented Generation (RAG) frameworks and their evolution into more advanced agent-based architectures within personalized settings to enhance user satisfaction. Building on this foundation, this survey systematically examines personalization across the three core stages of RAG: pre-retrieval, retrieval, and generation. Beyond RAG, we further extend its capabilities into the realm of Personalized LLM-based Agents, which enhance traditional RAG systems with agentic functionalities, including user understanding, personalized planning and execution, and dynamic generation. For both personalization in RAG and agent-based personalization, we provide formal definitions, conduct a comprehensive review of recent literature, and summarize key datasets and evaluation metrics. Additionally, we discuss fundamental challenges, limitations, and promising research directions in this evolving field. Relevant papers and resources are continuously updated at https://github.com/Applied-Machine-Learning-Lab/Awesome-Personalized-RAG-Agent.

• The paper presents a comprehensive exploration of personalization in AI, detailing the evolution from RAG frameworks to agent-based systems.
• By analyzing pre-retrieval, retrieval, and generation stages, it demonstrates techniques like query reformulation, indexing, and context-aware output generation.
• It outlines the transition to personalized LLM-based agents while addressing challenges in scalability, evaluation metrics, and privacy.

A Survey of Personalization: From RAG to Agent

Introduction

The paper "A Survey of Personalization: From RAG to Agent" (2504.10147) examines personalization in artificial intelligence, particularly focusing on the evolution from Retrieval-Augmented Generation (RAG) frameworks to advanced agent-based architectures, herein referred to as LLM-based Agents. These systems enhance user satisfaction by tailoring interactions to individual preferences at various stages of the pipeline, namely pre-retrieval, retrieval, and generation.

Personalization in RAG Frameworks

Pre-Retrieval Stage

Pre-retrieval involves enhancing user queries prior to retrieval actions, thereby refining outputs' relevance and quality. This stage integrates explicit user profiles, historical interactions, and persona-based simulations to reformulate queries. It utilizes techniques like query rewriting and expansion to adapt to user-specific contexts.

Figure 1: Overview of the personalized pre-retrieval stage.

Retrieval Stage

Personalized retrieval employs indexing techniques, both dense and sparse, to structure knowledge bases. It also leverages prompt-based retrieval strategies guided by user-specific instructions or interactions. This stage ensures the delivery of relevant, user-centered results, improving downstream outputs' personalization.

Figure 2: Overview of the personalized retrieval stage.

Generation Stage

Generation seamlessly integrates retrieved artifacts with personalized context, ensuring the alignment of outputs to user preferences. This stage employs two primary strategies: explicit personalization through context-aware prompts and implicit personalization via parameter tuning techniques like LoRA.

Figure 3: Overview of the personalized generation stage.

Transition to Personalized Agents

The convergence of personalized RAG processes with agent architectures forms "personalized RAG++," integrating deeper user-state tracking, adaptive tool usage, and context-aware generation. Personalized LLM-based agents dynamically incorporate user preferences into planning and execution phases, enabling contextually aligned, preference-sensitive outputs.

Figure 4: Overview of transition from personalized RAG to personalized agent.

Planning and Execution

Agents manage complex tasks leveraging advanced memory management techniques and real-time APIs, ensuring outcomes adhere to user-specific constraints. This transition emphasizes agents' capabilities to handle intricate interactions, dynamically aligning with user contexts.

Personalized Understanding and Interaction

Personalized understanding encompasses an agent's ability to decode user queries and preferences dynamically, enhancing role understanding. The integration of user-role joint frameworks further augments the agent’s adaptability across different contexts and interactions.

Challenges and Future Directions

The research identifies challenges in balancing system scalability with personalization complexity, optimizing evaluation metrics, and preserving privacy. Future work is directed towards lightweight frameworks, specialized evaluation metrics, and ethical system design to address these challenges effectively.

Conclusion

This survey synthesizes recent advancements in personalization within RAG frameworks and their convergence towards intelligent, personalized agents. By refining personalization from query inception through generation, it paves the way for nuanced, user-adaptive AI systems. Future directions include optimizing system scalability, advancing evaluation metrics, and ethical considerations, ensuring robust personalized interactions across diverse applications.