Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

The Oscars of AI Theater: A Survey on Role-Playing with Language Models

Nuo Chen, Yan Wang, Yang Deng, Jia Li

TL;DR

This survey analyzes role-playing with language models, distinguishing persona-based and character-based approaches and proposing a taxonomy around data, models-and-alignment, agent architecture, and evaluation. It reviews data sources, collection methods, and the granularity of role-related information; surveys foundation-model development and alignment strategies, including SFT, RLHF, ICL, and RAG. The paper also details RPLA architectures with memory, planning, and action modules, and surveys multifaceted evaluation methods, highlighting the need for robust, safe, and scalable assessment. It discusses major challenges—dynamic persona management, safety, hallucinations, memory efficiency, and multimodal integration—and outlines future directions for deeper alignment, better metrics, and lifelong learning in role-playing LLM systems.

Abstract

This survey explores the burgeoning field of role-playing with language models, focusing on their development from early persona-based models to advanced character-driven simulations facilitated by Large Language Models (LLMs). Initially confined to simple persona consistency due to limited model capabilities, role-playing tasks have now expanded to embrace complex character portrayals involving character consistency, behavioral alignment, and overall attractiveness. We provide a comprehensive taxonomy of the critical components in designing these systems, including data, models and alignment, agent architecture and evaluation. This survey not only outlines the current methodologies and challenges, such as managing dynamic personal profiles and achieving high-level persona consistency but also suggests avenues for future research in improving the depth and realism of role-playing applications. The goal is to guide future research by offering a structured overview of current methodologies and identifying potential areas for improvement. Related resources and papers are available at https://github.com/nuochenpku/Awesome-Role-Play-Papers.

The Oscars of AI Theater: A Survey on Role-Playing with Language Models

TL;DR

This survey analyzes role-playing with language models, distinguishing persona-based and character-based approaches and proposing a taxonomy around data, models-and-alignment, agent architecture, and evaluation. It reviews data sources, collection methods, and the granularity of role-related information; surveys foundation-model development and alignment strategies, including SFT, RLHF, ICL, and RAG. The paper also details RPLA architectures with memory, planning, and action modules, and surveys multifaceted evaluation methods, highlighting the need for robust, safe, and scalable assessment. It discusses major challenges—dynamic persona management, safety, hallucinations, memory efficiency, and multimodal integration—and outlines future directions for deeper alignment, better metrics, and lifelong learning in role-playing LLM systems.

Abstract

This survey explores the burgeoning field of role-playing with language models, focusing on their development from early persona-based models to advanced character-driven simulations facilitated by Large Language Models (LLMs). Initially confined to simple persona consistency due to limited model capabilities, role-playing tasks have now expanded to embrace complex character portrayals involving character consistency, behavioral alignment, and overall attractiveness. We provide a comprehensive taxonomy of the critical components in designing these systems, including data, models and alignment, agent architecture and evaluation. This survey not only outlines the current methodologies and challenges, such as managing dynamic personal profiles and achieving high-level persona consistency but also suggests avenues for future research in improving the depth and realism of role-playing applications. The goal is to guide future research by offering a structured overview of current methodologies and identifying potential areas for improvement. Related resources and papers are available at https://github.com/nuochenpku/Awesome-Role-Play-Papers.
Paper Structure (50 sections, 7 figures, 1 table)

This paper contains 50 sections, 7 figures, 1 table.

Table of Contents

  1. Introduction
  2. Background
  3. PLMs and LLMs
  4. Key Components
  5. Data
  6. Data Objectives
  7. Persona-based Role-Playing Datasets
  8. Interaction Collection
  9. Role-related Information.
  10. Character-based Role-Playing Datasets
  11. Interaction Collection
  12. Role-related Information.
  13. Models and Alignment
  14. Foundation Models
  15. Non-pretrained models and PLM.
  16. ...and 35 more sections

Figures (7)

  • Figure 1: Key components in role-playing with language models.
  • Figure 2: The main content flow and categorization of Data Section.
  • Figure 3: Typical (a) persona-based and (b) character-based role-related information in role-playing datasets. The given examples (a) are from Persona-Chat zhang-etal-2018-personalizing and PersonalDialog DBLP:journals/corr/abs-1901-09672. Example (b) is collected from HPD chen2023large.
  • Figure 4: The main content flow and categorization of Models and Alignment.
  • Figure 5: The main content flow and categorization of Agent Architecture.
  • ...and 2 more figures