Home

Awesome

Two Tales of Persona in LLMs: <br> A Survey of Role-Playing and Personalization

Static Badge GitHub Repo stars GitHub last commit

<br> <p align="center"> <img src="figures/Overview.png" alt="Overview" width="40%" height="auto"> </p>

Introduction

This is the official repository of the paper "Two Tales of Persona in LLMs: A Survey of Role-Playing and Personalization", EMNLP 2024 Findings.

The concept of persona, originally adopted in dialogue literature, has re-surged as a promising framework for tailoring large language models (LLMs) to specific context (e.g., personalized search, LLM-as-a-judge). However, the growing research on leveraging persona in LLMs is relatively disorganized and lacks a systematic taxonomy. To close the gap, we present a comprehensive survey to categorize the current state of the field. We identify two lines of research, namely (1) LLM Role-Playing, where personas are assigned to LLMs, and (2) LLM Personalization, where LLMs take care of user personas. Additionally, we introduce existing methods for LLM personality evaluation. To the best of our knowledge, we present the first survey for role-playing and personalization in LLMs under the unified view of persona.

We continuously maintain this paper collection to foster future endeavors.

<div style="display: flex; align-items: center;"> <div style="flex: 1;">

News

Table of Contents

<h2 id="llm-role-play-adapt-to-environment">🙆‍♀️ LLM Role-Play (Adapt to Environment)</h2>

LLMs are tasked to play the assigned personas (i.e., roles) and act accordance to environmental feedback.

The key aspect is how LLMs adapt to defined environments.

<br> <p align="center"> <a href="."> <img src="figures/llm-role-playing.png" alt="LLM role-playing" width="90%" height="auto"> </a> </p> <h3 id="role-playing-workshops">💼 Workshops</h3>
DateWorkshopWebsite Link
2405LLMAgent @ ICLRICLR 2024 Workshop on Large Language Model (LLM) Agents
2405Agent Workshop @ CMUCMU Agent Workshop 2024
<h3 id="environments">🌎 Environments</h3> <h4 id="software-development">💻 Software Development</h4>
DateAuthorsVenuePaper
2308Hong et al.ICLRMetaGPT: Meta Programming for Multi-Agent Collaborative Framework
2307Qian et al.arXivCommunicative agents for software development
2305Dong et al.TOSEMSelf-collaboration code generation via chatgpt
2107Chen et al.arXivEvaluating large language models trained on code
<h4 id="web">🌐 Web</h4>
DateAuthorsVenuePaper
2404Liu et al.arXivVisualWebBench: How Far Have Multimodal LLMs Evolved in Web Page Understanding and Grounding?
2401Zheng et al.LLMAgent @ ICLRGPT-4V(ision) is a Generalist Web Agent, if Grounded
2401Koh et al.LLMAgent @ ICLRVisualWebArena: Evaluating Multimodal Agents on Realistic Visual Web Tasks
2401Cheng et al.LLMAgent @ ICLRSeeClick: Harnessing GUI Grounding for Advanced Visual GUI Agents
2312Gur et al.EMNLPUnderstanding HTML with Large Language Models
2312Hong et al.arXivCogAgent: A Visual Language Model for GUI Agents
2307Gur et al.ICLRA Real-World WebAgent with Planning, Long Context Understanding, and Program Synthesis
2307Zhou et al.ICLRWebArena: A Realistic Web Environment for Building Autonomous Agents
2306Deng et al.NeurIPSMind2web: Towards a generalist agent for the web
2303Kim et al.NeurIPSLanguage Models can Solve Computer Tasks
<h4 id="game">🎮 Game</h4>
DateAuthorsVenuePaper
2310Wang et al.EMNLPHumanoid Agents: Platform for Simulating Human-like Generative Agents
2305Wang et al.TMLRVoyager: An open-ended embodied agent with large language models
2305Fu et al.arXivImproving Language Model Negotiation with Self-Play and In-Context Learning from AI Feedback
2304Park et al.UISTGenerative agents: Interactive simulacra of human behavior
<h4 id="medical-application">🏥 Medical Application</h4>
DateAuthorsVenuePaper
2312Kwon et al.AAAILarge Language Models are Clinical Reasoners: Reasoning-Aware Diagnosis Framework with Prompt-Generated Rationales
2311Tang et al.arXivMedagents: Large language models as collaborators for zero-shot medical reasoning
2307Wu et al.ICLRLarge Language Models Perform Diagnostic Reasoning
2207Liévin et al.arXivCan large language models reason about medical questions?
<h4 id="llm-as-evaluators">🧑‍⚖️ LLM as Evaluators</h4>
DateAuthorsVenuePaper
2308Chan et al.ICLRChatEval: Towards Better LLM-based Evaluators through Multi-Agent Debate
2303Wu et al.NLPCCLarge Language Models are Diverse Role-Players for Summarization Evaluation
<h4 id="general-framework">📦 General Framework</h4>
DateAuthorsVenuePaper
2405Ahn, et alACL FindingsTimeChara: Evaluating Point-in-Time Character Hallucination of Role-Playing Large Language Models
2308Chen et al.ICLRAgentVerse: Facilitating Multi-Agent Collaboration and Exploring Emergent Behaviors
2307Wang et al.NAACLUnleashing the Emergent Cognitive Synergy in Large Language Models: A Task-Solving Agent through Multi-Persona Self-Collaboration
2303Li et al.NeurIPSCAMEL: Communicative Agents for "Mind" Exploration of Large Language Model Society
<h3 id="agentic-interactions">🤖 Interaction & Behaviors</h3> <h4 id="schemas">📊 Schemas</h4> <h5 id="single-agent">👤 Single-Agent</h5>
DateAuthorsVenuePaper
2401Cheng et al.LLMAgent @ ICLRSeeClick: Harnessing GUI Grounding for Advanced Visual GUI Agents
2401Zheng et al.LLMAgent @ ICLRGPT-4V(ision) is a Generalist Web Agent, if Grounded
2312Hong et al.arXivCogAgent: A Visual Language Model for GUI Agents
2305Wang et al.TMLRVoyager: An open-ended embodied agent with large language models
<h5 id="multi-agent">👥 Multi-Agent</h5>
DateAuthorsVenuePaper
2311Tang et al.arXivMedagents: Large language models as collaborators for zero-shot medical reasoning
2308Chen et al.ICLRAgentVerse: Facilitating Multi-Agent Collaboration and Exploring Emergent Behaviors
2308Hong et al.ICLRMetaGPT: Meta Programming for Multi-Agent Collaborative Framework
2308Chan et al.ICLRChatEval: Towards Better LLM-based Evaluators through Multi-Agent Debate
2307Qian et al.arXivCommunicative agents for software development
2305Fu et al.arXivImproving Language Model Negotiation with Self-Play and In-Context Learning from AI Feedback
<h4 id="emergent-behaviors">💡 Emergent Behaviors</h4>
DateAuthorsVenuePaper
2311Tang et al.arXivMedagents: Large language models as collaborators for zero-shot medical reasoning
2308Chen et al.ICLRAgentVerse: Facilitating Multi-Agent Collaboration and Exploring Emergent Behaviors
2307Wang et al.NAACLUnleashing the Emergent Cognitive Synergy in Large Language Models: A Task-Solving Agent through Multi-Persona Self-Collaboration
2305Fu et al.arXivImproving Language Model Negotiation with Self-Play and In-Context Learning from AI Feedback
2303Li et al.NeurIPSCAMEL: Communicative Agents for "Mind" Exploration of Large Language Model Society
<h2 id="llm-personalization-adapt-to-user">🙆‍♂️ LLM Personalization (Adapt to User)</h2>

LLMs are tasked to take care of users’ personas (e.g., background information, or historical behaviors) to meet customized needs.

The key aspect is how LLMs adapt to distinct users.

<p align="center"> <a href="."> <img src="figures/llm-personalization.png" alt="LLM personalization" width="90%" height="auto"> </a> </p> <h3 id="personalization-workshops">💼 Workshops & Competitions</h3>
DateAuthorsVenuePaper
2403Deshpande et al.PERSONALIZE @ EACLProceedings of the 1st Workshop on Personalization of Generative AI Systems (PERSONALIZE 2024)
2310Chen et al.Personalized Generative AI @ CIKMThe First Workshop on Personalized Generative AI @ CIKM 2023: Personalization Meets Large Language Models
1902Dinan et al.ConvAI2 @ NeurIPSThe Second Conversational Intelligence Challenge (ConvAI2)
1808Yusupov et al.ConvAI @ NeurIPSNIPS Conversational Intelligence Challenge 2017 Winner System: Skill-based Conversational Agent with Supervised Dialog Manager
<h3 id="tasks">📌 Tasks</h3> <h4 id="personalized-dialogue">💬 Personalized Dialogue</h4> <h5 id="tod-modeling">🔧 ToD Modeling</h5>

LLMs Era

DateAuthorsVenuePaper
2305Yang et al.EMNLPRefGPT: Dialogue Generation of GPT, by GPT, and for GPT
2302Li et al.NeurIPSGuiding large language models via directional stimulus prompting
2005Hosseini-Asl et al.NeurIPSA Simple Language Model for Task-Oriented Dialogue
<details> <summary style="color:#FFB6C1">Comprehensive Paper List</summary>
DateAuthorsVenuePaper
2312Xu et al.EMNLPBaize: An Open-Source Chat Model with Parameter-Efficient Tuning on Self-Chat Data
2309Hu et al.arXivEnhancing Large Language Model Induced Task-Oriented Dialogue Systems Through Look-Forward Motivated Goals
2308Wu et al.SIGDIALDiactTOD: Learning Generalizable Latent Dialogue Acts for Controllable Task-Oriented Dialogue Systems
2305Yang et al.EMNLPRefGPT: Dialogue Generation of GPT, by GPT, and for GPT
2305Bang et al.ACLTask-Optimized Adapters for an End-to-End Task-Oriented Dialogue System
2304Ashby et al.CHIPersonalized Quest and Dialogue Generation in Role-Playing Games: A Knowledge Graph and Language Model-based Approach
2304Hudevcek et al.SIGDIALAre Large Language Models All You Need for Task-Oriented Dialogue?
2302Li et al.NeurIPSGuiding large language models via directional stimulus prompting
2302Feng et al.ICLRFantastic rewards and how to tame them: A case study on reward learning for task-oriented dialogue systems
2210Huryn et al.COLINGAutomatic Generation of Large-scale Multi-turn Dialogues from Reddit
2108Peng et al.TACLSoloist: Building Task Bots at Scale with Transfer Learning and Machine Teaching
2012Yang et al.AAAIUBAR: Towards Fully End-to-End Task-Oriented Dialog Systems with GPT-2
2008Madotto et al.arXivLanguage models as few-shot learner for task-oriented dialogue systems
2005Hosseini-Asl et al.NeurIPSA Simple Language Model for Task-Oriented Dialogue
</details>

Pre-LLMs Era

DateAuthorsVenuePaper
2006Jianhong Wang et al.ICLRModelling hierarchical structure between dialogue policy and natural language generator with option framework for task-oriented dialogue system
1606N. Mrksic et al.ACLNeural Belief Tracker: Data-Driven Dialogue State Tracking
1506Alessandro Sordoni et al.NAACLA Neural Network Approach to Context-Sensitive Generation of Conversational Responses
<details> <summary style="color:#FFB6C1">Comprehensive Paper List</summary>
DateAuthorsVenuePaper
2105Sun et al.SIGIRSimulating user satisfaction for the evaluation of task-oriented dialogue systems
2006Wang et al.ICLRModelling hierarchical structure between dialogue policy and natural language generator with option framework for task-oriented dialogue system
2003Yang et al.IEEEMultitask Learning and Reinforcement Learning for Personalized Dialog Generation: An Empirical Study
1912HuangAAAIMALA: Cross-Domain Dialogue Generation with Action Learning
1910Zhang et al.*SEMFind or Classify? Dual Strategy for Slot-Value Predictions on Multi-Domain Dialog State Tracking
1905Wu et al.ACLTransferable Multi-Domain State Generator for Task-Oriented Dialogue Systems
1807Lei et al.ACLSequicity: Simplifying Task-oriented Dialogue Systems with Single Sequence-to-Sequence Architectures
1804Liu et al.NAACLDialogue Learning with Human Teaching and Feedback in End-to-End Trainable Task-Oriented Dialogue Systems
1712Rastogi et al.IEEEScalable Multi-Domain Dialogue State Tracking
1709Wu et al.AAAIStarSpace: Embed all the things!
1606Miller et al.EMNLPKey-Value Memory Networks for Directly Reading Documents
1606Mrksic et al.ACLNeural Belief Tracker: Data-Driven Dialogue State Tracking
1506Sordoni et al.NAACLA Neural Network Approach to Context-Sensitive Generation of Conversational Responses
</details> <h5 id="user-persona-modeling">📝 User Persona Modeling</h5>
DateAuthorsVenuePaper
2405HanACLPSYDIAL: Personality-based Synthetic Dialogue Generation Using Large Language Models
2307Tang et al.ACLEnhancing Personalized Dialogue Generation with Contrastive Latent Variables: Combining Sparse and Dense Persona
1807Zhang et al.ACLPersonalizing Dialogue Agents: I have a dog, do you have pets too?
<details> <summary style="color:#FFB6C1">Comprehensive Paper List</summary>
DateAuthorsVenuePaper
2405HanACLPSYDIAL: Personality-based Synthetic Dialogue Generation Using Large Language Models
2401Lotfi et al.IEEEPersonalityChat: Conversation Distillation for Personalized Dialog Modeling with Facts and Traits
2401Kim et al.EACLCommonsense-augmented Memory Construction and Management in Long-term Conversations via Context-aware Persona Refinement
2308Tu et al.arXivCharacterChat: Learning towards Conversational AI with Personalized Social Support
2307Tang et al.ACLEnhancing Personalized Dialogue Generation with Contrastive Latent Variables: Combining Sparse and Dense Persona
2307Ahn, et alACLMPCHAT: Towards Multimodal Persona-Grounded Conversation
2011Zhong et al.EMNLPTowards Persona-Based Empathetic Conversational Models
2007Wu et al.ACLGuiding Variational Response Generator to Exploit Persona
2007Liu et al.ACLYou Impress Me: Dialogue Generation via Mutual Persona Perception
1911Zheng et al.AAAIA Pre-Training Based Personalized Dialogue Generation Model with Persona-Sparse Data
1911Song et al.AAAIGenerating Persona Consistent Dialogues by Exploiting Natural Language Inference
1807Zhang et al.ACLPersonalizing Dialogue Agents: I have a dog, do you have pets too?
</details> <h4 id="recommendation-system">🛒 Recommendation System</h4>
DateAuthorsVenuePaper
2305Yang et al.arXivPALR: Personalization Aware LLMs for Recommendation
2304Wang et al.arXivZero-Shot Next-Item Recommendation using Large Pretrained Language Models
2108Li et al.ACLPersonalized Transformer for Explainable Recommendation
<details> <summary><span style="color:#FFB6C1">Comprehensive Paper List</span></summary>
DateAuthorsVenuePaper
2405Hu et al.WWWEnhancing sequential recommendation via llm-based semantic embedding learning
2311Chen et al.arXivA Survey on Large Language Models for Personalized and Explainable Recommendations
2308Wang et al.arXivRecMind: Large Language Model Powered Agent For Recommendation
2308Chu et al.arXivLeveraging Large Language Models for Pre-trained Recommender Systems
2306Li et al.arXivPrompt Tuning Large Language Models on Personalized Aspect Extraction for Recommendations
2305Yang et al.arXivPALR: Personalization Aware LLMs for Recommendation
2305Zhang et al.arXivRecommendation as Instruction Following: A Large Language Model Empowered Recommendation Approach
2304Wang et al.arXivZero-Shot Next-Item Recommendation using Large Pretrained Language Models
2208Chen et al.KDDPersonalized Chit-Chat Generation for Recommendation Using External Chat Corpora
2202Li et al.TOISPersonalized Prompt Learning for Explainable Recommendation
2108Li et al.ACLPersonalized Transformer for Explainable Recommendation
</details> <h4 id="personalized-search">🔍 Personalized Search</h4>
DateAuthorsVenuePaper
2405Zhou et al.WWWCognitive personalized search integrating large language models with an efficient memory mechanism
2405Baek et al.WWWKnowledge-Augmented Large Language Models for Personalized Contextual Query Suggestion
2405SalemiarXivUnified ranking for multiple retrieval-augmented large language models
2402Sharma et al.CHIGenerative echo chamber? effects of llm-powered search systems on diverse information seeking
2307Eleni et al.arXivComparing Traditional and LLM-based Search for Consumer Choice: A Randomized Experiment
2307Ziems et al.ACLLarge Language Models are Built-in Autoregressive Search Engines
2107Zhou et al.SIGIRGroup based Personalized Search by Integrating Search Behaviour and Friend Network
<h4 id="personalized-healthcare">🩺 Personalized Healthcare</h4>
DateAuthorsVenuePaper
2402Abbasian et al.arXivKnowledge-Infused LLM-Powered Conversational Health Agent: A Case Study for Diabetes Patients
2402Jin et al.arXivHealth-LLM: Personalized Retrieval-Augmented Disease Prediction System
2310Abbasian et al.arXivConversational Health Agents: A Personalized LLM-Powered Agent Framework
2309Zhang et al.arXivLLM-based Medical Assistant Personalization with Short- and Long-Term Memory Coordination
<h4 id="personalized-education">📚 Personalized Education</h4>
DateAuthorsVenuePaper
2403Park et al.CHIEmpowering personalized learning through a conversation-based tutoring system with student modeling
2308Dan et al.arXivEduchat: A large-scale language model-based chatbot system for intelligent education
2307Shehata et al.BEA @ ACLEnhancing Video-based Learning Using Knowledge Tracing: Personalizing Students’ Learning Experience with ORBITS
<h3 id="methods">🛠️ Methods</h3> <h4 id="fine-tuning">🎛️ Fine-Tuning</h4>
DateAuthorsVenuePaper
2403Mondal et al.EACLPresentations by the Humans and For the Humans: Harnessing LLMs for Generating Persona-Aware Slides from Documents
2402Li et al.arXivPersonalized Language Modeling from Personalized Human Feedback
2402Tan et al.arXivDemocratizing Large Language Models via Personalized Parameter-Efficient Fine-tuning
2312Hwang et al.arXivPromptable Behaviors: Personalizing Multi-Objective Rewards from Human Preferences
2312Shea et al.EMNLPBuilding Persona Consistent Dialogue Agents with Offline Reinforcement Learning
2311Qin et al.arXivEnabling on-device large language model personalization with self-supervised data selection and synthesis
2310Jang et al.arXivPersonalized large language model alignment via post-hoc parameter merging
2303Kirk et al.arXivPersonalisation within bounds: A risk taxonomy and policy framework for the alignment of large language models with personalised feedback
<h4 id="retrieval-augmentation">🔗 Retrieval Augmentation</h4>
DateAuthorsVenuePaper
2404Zhang et al.arXivPersonalized LLM Response Generation with Parameterized Memory Injection
2403Zhong et al.AAAIMemoryBank: Enhancing Large Language Models with Long-Term Memory
2402Sun et al.arXivPersona-DB: Efficient Large Language Model Personalization for Response Prediction with Collaborative Data Refinement
2402Tan et al.arXivDemocratizing Large Language Models via Personalized Parameter-Efficient Fine-tuning
2205Fu et al.ACLThere Are a Thousand Hamlets in a Thousand People’s Eyes: Enhancing Knowledge-grounded Dialogue with Personal Memory
2106Wu et al.NAACLPersonalized Response Generation via Generative Split Memory Network
<h4 id="prompting">✍️ Prompting</h4> <h5 id="vanilla-personalized-prompt">📄 Vanilla Personalized Prompt</h5>
DateAuthorsVenuePaper
2305Dai et al.RecSysUncovering ChatGPT’s Capabilities in Recommender Systems
2305Christakopoulou et al.arXivLarge Language Models for User Interest Journeys
2305Zhiyuli et al.arXivBookGPT: A General Framework for Book Recommendation Empowered by Large Language Model
<h5 id="retrieval-augmented-personalized-prompt">🔦 Retrieval-Augmented Personalized Prompt</h5>
DateAuthorsVenuePaper
2311Mysore et al.arXivPEARL: Personalizing Large Language Model Writing Assistants with Generation-Calibrated Retrievers
2308Li et al.arXivTeach LLMs to Personalize -- An Approach inspired by Writing Education
2304Salemi et al.arXivLaMP: When Large Language Models Meet Personalization
<h5 id="profile-augmented-prompt">📂 Profile-Augmented Prompt</h5>
DateAuthorsVenuePaper
2405Li et al.WWWLearning to Rewrite Prompts for Personalized Text Generation
2310Richardson et al.arXivIntegrating Summarization and Retrieval for Enhanced Personalization via Large Language Models
2305Liu et al.WSDMONCE: Boosting Content-based Recommendation with Both Open- and Closed-source Large Language Models
<h2 id="llm-personality-evaluation">🧐 LLM Personality Evaluation</h2>
DateAuthorsVenuePaper
2401Huang et al.ICLROn the Humanity of Conversational AI: Evaluating the Psychological Portrayal of LLMs
2309Jiang et al.NeurIPSEvaluating and inducing personality in pre-trained language models
2307Fang et al.ACLOn Text-based Personality Computing: Challenges and Future Directions
<details> <summary><span style="color:#FFB6C1">Comprehensive Paper List</span></summary>
DateAuthorsVenuePaper
2403Sorokovikova et al.PERSONALIZE @ EACLLLMs Simulate Big5 Personality Traits: Further Evidence
2403Frisch et al.PERSONALIZE @ EACLLLM Agents in Interaction: Measuring Personality Consistency and Linguistic Alignment in Interacting Populations of Large Language Models
2402Song et al.arXivIdentifying Multiple Personalities in Large Language Models with External Evaluation
2402Song et al.arXivIdentifying Multiple Personalities in Large Language Models with External Evaluation
2402Yang et al.arXivLLM Agents for Psychology: A Study on Gamified Assessments
2401Huang et al.ICLROn the Humanity of Conversational AI: Evaluating the Psychological Portrayal of LLMs
2312Rao et al.EMNLPCan ChatGPT Assess Human Personalities? A General Evaluation Framework
2311Dorner et al.SoLaR @ NeurIPSDo personality tests generalize to large language models?
2310Wang et al.arXivInCharacter: Evaluating Personality Fidelity in Role-Playing Agents through Psychological Interviews
2309Jiang et al.NeurIPSEvaluating and inducing personality in pre-trained language models
2307Pan et al.arXivDo LLMs Possess a Personality? Making the MBTI Test an Amazing Evaluation for Large Language Models
2307Fang et al.ACLOn Text-based Personality Computing: Challenges and Future Directions
2307Ji et al.arXivIs ChatGPT a Good Personality Recognizer? A Preliminary Study
2305Jiang et al.arXivPersonallm: Investigating the ability of large language models to express big five personality traits
</details> <h2 id="how-to-contribute">🌱 How to contribute</h2>

:sparkles: Welcome to contribute to this reading list via :memo: Issues using the following format.

DateAuthorsVenuePaper
1706Vaswani, et alNeurIPSAttention Is All You Need
</div> </div> <h2 id="citation">🔖 Citation</h2>

📚 If you find our survey beneficial for your research, please kindly cite our paper :-)

@misc{tseng2024talespersonallmssurvey,
  title={Two Tales of Persona in LLMs: A Survey of Role-Playing and Personalization},
  author={Yu-Min Tseng and Yu-Chao Huang and Teng-Yun Hsiao and Wei-Lin Chen and Chao-Wei Huang and Yu Meng and Yun-Nung Chen},
  year={2024},
  eprint={2406.01171},
  archivePrefix={arXiv},
  primaryClass={cs.CL},
  url={https://arxiv.org/abs/2406.01171},
}
<h2 id="authors">🖌️ Authors</h2>

Yu-Min Tseng*, Yu-Chao Huang*, Teng-Yun Hsiao*, Wei-Lin Chen*, Chao-Wei Huang, Yu Meng, Yun-Nung Chen.

(* Equal Contribution.) (Acknowlegement: Yu-Ching Hsu, Jia-Yin Foo.)

<div align="center"> <img src="figures/ntu-logo.png" width="20%" /> <img src="figures/uva-logo.png" width="25%" /> </div>