Identifying and Mitigating Bottlenecks in Role-Playing Agents: A Systematic Study of Disentangling Character Profile Axes
On the map: LLM agents and tool use → Social simulation and multi-agent debate
Abstract (arXiv)
While Large Language Model (LLM) role-playing agents have advanced rapidly, it remains unclear which profile elements genuinely drive role-playing quality. To bridge this gap, we introduce a systematic diagnostic framework that disentangles the impact of character profiles along three axes: Familiarity (Known vs. Unknown), Structure (Structured vs. Unstructured), and Disposition (Moral vs. Immoral). Utilizing a unified hierarchical schema (5 dimensions, 28 fields), we construct a controlled dataset of 211 personas and evaluate five LLMs on both single- and multi-turn interactions. Our results reveal a striking asymmetry: Familiarity and Structure show negligible impact, while Disposition produces large, consistent performance degradation for immoral characters across all conditions. Further analyses suggest that the Moral--Immoral gap is amplified by post-SFT alignment, and that this degradation varies substantially across profile attributes. To mitigate this bottleneck, we propose Field-Aware Contrastive Decoding (FACD), a training-free strategy that amplifies suppressed disposition-sensitive signals, significantly closing the performance gap without sacrificing moral-character performance.
Key facts. Cited by 2 papers in the map (data through 7 Oct 2026). Placed in Social simulation and multi-agent debate, one of 16 directions in LLM agents and tool use. Builds on 35 of its 63 references that are in the map. First posted on arXiv 8 Jan 2026.
Builds on
35 of 63 references.
Closest ancestors · 7
Same direction, newest first
- 2024Examining Identity Drift in Conversations of LLM Agents1 Dec
- 2024Emotional RAG: Enhancing Role-Playing Agents through Emotional Retrieval30 Oct
- 2024Identity-Driven Hierarchical Role-Playing Agents28 Jul
- 2024PersonaGym: Evaluating Persona Agents and LLMs25 Jul
- 2024Crafting Customisable Characters with LLMs: A Persona-Driven Role-Playing Agent Framework25 Jun
- 2023Character-LLM: A Trainable Agent for Role-Playing16 Oct
- 2023Generative Agents: Interactive Simulacra of Human Behavior7 Apr
Other references · 28
- 2022Training language models to follow instructions with human feedbackcited by 15,712
- 2023Direct Preference Optimization: Your Language Model is Secretly a Reward Modelcited by 7,916
- 2024The FineWeb Datasets: Decanting the Web for the Finest Text Data at Scalecited by 1,104
- 2019BERT Rediscovers the Classical NLP Pipelinecited by 1,046
- 2023Towards Understanding Sycophancy in Language Modelscited by 988
- 2019Attention is not Explanationcited by 916
- 2023Quantifying Language Models' Sensitivity to Spurious Features in Prompt Design or: How I learned to start worrying about prompt formattingcited by 778
- 2020Quantifying Attention Flow in Transformerscited by 759
- 2020Transformer Interpretability Beyond Attention Visualizationcited by 620
- 2022Contrastive Decoding: Open-ended Text Generation as Optimizationcited by 598
- 2023FlexGen: High-Throughput Generative Inference of Large Language Models with a Single GPUcited by 567
- 2024Knowledge Conflicts for LLMs: A Surveycited by 274
16 more
- 2023PersonaLLM: Investigating the Ability of Large Language Models to Express Personality Traitscited by 229
- 2023SARATHI: Efficient LLM Inference by Piggybacking Decodes with Chunked Prefillscited by 187
- 2023InCharacter: Evaluating Personality Fidelity in Role-Playing Agents through Psychological Interviewscited by 177
- 2024Does Prompt Formatting Have Any Impact on LLM Performance?cited by 132
- 2025Safety Tax: Safety Alignment Makes Your Large Reasoning Models Less Reasonablecited by 120
- 2024Large Language Models are Superpositions of All Characters: Attaining Arbitrary Role-play via Self-Alignmentcited by 104
- 2024Evaluating Character Understanding of Large Language Models via Character Profiling from Fictional Workscited by 38
- 2024CharacterBox: Evaluating the Role-Playing Capabilities of LLMs in Text-Based Virtual Worldscited by 38
- 2025OpenCharacter: Training Customizable Role-Playing LLMs with Large-Scale Synthetic Personascited by 25
- 2024MoralBERT: A Fine-Tuned Language Model for Capturing Moral Values in Social Discussionscited by 23
- 2025From Personas to Talks: Revisiting the Impact of Personas on LLM-Synthesized Emotional Support Conversationscited by 23
- 2024Mitigating Hallucination in Fictional Character Role-Playcited by 17
- 2023"Merge Conflicts!" Exploring the Impacts of External Distractors to Parametric Knowledge Graphscited by 11
- 2025Too Good to be Bad: On the Failure of LLMs to Role-Play Villainscited by 10
- 2024Revealing and Mitigating the Challenge of Detecting Character Knowledge Errors in LLM Role-Playingcited by 9
- 2025Exploring Persona Sentiment Sensitivity in Personalized Dialogue Generationcited by 3
Built on by
Citing papers per year: 2026: 2.
2 papers.
Similar, not cited
Close in wording, no citation either way
- 2026Rethinking Role-Playing Evaluation: Anonymous Benchmarking and a Systematic Study of Personality Effects
- 2025Test-Time-Matching: Decouple Personality, Memory, and Linguistic Style in LLM-based Role-Playing Language Agent
- 2024Quantifying and Optimizing Global Faithfulness in Persona-driven Role-playing
- 2025Enhancing Persona Consistency for LLMs' Role-Playing using Persona-Aware Contrastive Learning
- 2026RoleCDE:Benchmarking and Mitigating Role-Alignment Trade-offs in Role-Playing Agents
- 2026Cognitive Enhancement: Rethinking the Necessity of Role-Playing for Large Language Models
- 2025DPRF: A Generalizable Dynamic Persona Refinement Framework for Optimizing Behavior Alignment Between Personalized LLM Role-Playing Agents and Humans
- 2025Operation Veja: Fixing Fundamental Concepts Missing from Modern Roleplaying Training Paradigms