Research map / Large language models / Language, safety and interpretability
RPM: Reasoning-Level Personalization for Black-Box Large Language Models
Jieyong Kim, Tongyoung Kim, SooJin Yoon, Jaehyung Kim, Dongha Lee
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 RPM reasoning-level personalization vs best response-level baselines | GPT-4o-mini · LaMP-2 | accuracy: 0.526 -> 0.561 | Improved |
| add RPM vs best baseline | GPT-4o-mini · LaMP-3 | MAE: 0.312 -> 0.259 | Improved |
| add chain-of-thought prompting to baselines | ICL, RAG, HYDRA, Fermi · LaMP-2 | accuracy: 0.526 -> 0.493 (RAG) | Hurt |
| add personalized reasoning generation for the target query (ablation) | RPM · LaMP-3 | MAE: 0.305 -> 0.259 | Improved |
| add personalized reasoning generation for the target query (ablation) | RPM · GOQA | accuracy: 0.820 -> 0.852 | Improved |
| use generic CoT reasoning examples instead of personalized reasoning (ablation) | few-shot prompting · LaMP-3 | MAE: 0.274 -> 0.385 | Hurt |
| add personalized reasoning paths with features and factors (ablation) | few-shot with features/factors · GOQA | accuracy: 0.806 -> 0.852 | Improved |
| add user factors and target features to zero-shot input (ablation) | zero-shot · LaMP-3 | MAE: 0.361 -> 0.287 | Improved |
| add user factors and target features to zero-shot input (ablation) | zero-shot · LaMP-5 | ROUGE-1: 0.446 -> 0.427 | Hurt |
| retrieve with extracted features instead of raw query (ablation) | RPM over reasoning-augmented history · LaMP-2 | accuracy: 0.542 -> 0.561 | Improved |
| retrieve with extracted features instead of raw query (ablation) | RPM over reasoning-augmented history · LaMP-3 | MAE: 0.272 -> 0.259 | Improved |
| retrieve with features from raw history instead of reasoning-augmented history (ablation) | RPM · GOQA | accuracy: 0.809 -> 0.852 | Improved |
Closest papers by meaning
- HYDRA: Model Factorization Framework for Black-Box LLM Personalization (NeurIPS 2024 · similarity 0.89)
- Rethinking Personalization in Large Language Models at the Token Level (ICML 2026 · similarity 0.89)
- How Does Personalized Memory Shape LLM Behavior? Benchmarking Rational Preference Utilization in Personalized Assistants (NeurIPS 2026 · similarity 0.88)
- PersonaVLM: Long-Term Personalized Multimodal LLMs (CVPR 2026 · similarity 0.88)
- Personalize-then-Store: Benchmarking and Learning Personalized Memory for Long-Horizon Agents (NeurIPS 2026 · similarity 0.86)
- Think-While-Generating: On-the-Fly Reasoning for Personalized Long-Form Generation (ICLR 2026 · similarity 0.86)
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.
