Research map / Large language models / Alignment and preferences
Enhancing Persona Following at Decoding Time via Dynamic Importance Estimation for Role-Playing Agents
Yuxin Liu, Mingye Zhu, Siyuan Liu, Bo Hu, Lei Zhang
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add reward normalization to PDD (ablation) | PDD · CharacterEval (Qwen2.5-7B) | win rate (%): 38 -> 42 | Improved |
| add reward normalization to PDD (ablation) | PDD · CharacterEval (LLaMA-3-8B) | win rate (%): 32 -> 40 | Improved |
| add dynamic persona importance decoding (PDD) vs persona prompting | Persona Prompting · CharacterEval average (Qwen) | average score: 2.83 -> 2.85 | Improved |
| use PDD instead of ICL on LLaMA-3-8B | ICL · CharacterEval average | average score: 2.75 -> 2.81 | Improved |
| PDD vs PP win rate on LLaMA | PP · CharacterEval | win rate: 39.1 win / 31.3 lose | Mixed |
| PDD vs NPTI on Neuroticism | NPTI · PersonalityBench Neuroticism (LLaMA) | trait intensity: 4.82 -> 4.13 | Hurt |
| add PDD over best baseline on Big-Five average | NPTI/PP · PersonalityBench (Qwen) | average: 4.53 -> 4.57 | Improved |
| increase number of persona attributes aligned (ablation) | PDD · CharacterEval | quality | Mixed |
| degrade quality of G with top-k sampling (ablation) | PIE · CharacterEval | importance metrics | No clear effect |
Closest papers by meaning
- Persona Vectors: Monitoring and Controlling Character Traits in Language Models (NeurIPS 2026 · similarity 0.88)
- LLM Generated Persona is a Promise with a Catch (NeurIPS 2025 · similarity 0.88)
- PersonaManifold: Revealing and Exploiting Curved Geometry in LLM Persona Representations (NeurIPS 2026 · similarity 0.87)
- Quantifying and Optimizing Global Faithfulness in Persona-driven Role-playing (NeurIPS 2024 · similarity 0.87)
- Persona-Model Collapse in Emergent Misalignment (NeurIPS 2026 · similarity 0.87)
- Persona Generators: Generating Diverse Synthetic Personas for Arbitrary Contexts (NeurIPS 2026 · similarity 0.87)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
