Personalized Alignment in the Era of LLMs: A Review of Methodologies, Datasets, Evaluation, and Applications

With the rapid development of Large Language Models (LLMs), there is an increasing demand to adapt their behavior to individual user preferences, values, and contexts. However, the prevailing One-Size-Fits-All paradigm, which employs a single unified model to serve all users, falls short in achieving effective personalized alignment. To improve this, a growing body of research has introduced the One-Size-Fits-One approach, which customizes model services for specific user groups, thereby enabling more precise and effective personalization. In this paper, we present a comprehensive survey of personalized alignment in LLMs, aiming to provide a structured overview of the current state of research and future directions. To this end, we propose a unified classification framework to systematically categorize existing methods, datasets, evaluation, and applications. Based on the categories defined in our framework, we systematically review and summarize representative works, analyze their strengths and limitations, and discuss emerging trends and future directions in the field of personalized LLMs. Our goal is to lay the foundation for future research and practice in the design, implementation, and evaluation of personalized LLMs. Relevant papers and resources are continuously updated at our GitHub repository1.