Research map / Large language models / Alignment and preferences
NextQuill: Causal Preference Modeling for Enhancing LLM Personalization
Xiaoyan Zhao, Juntao You, Yang Zhang, Wenjie Wang, Hong Cheng, Fuli Feng, See-Kiong Ng, Tat-Seng Chua
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 causal preference alignment (NextQuill) | ContextSFT · Book Review | ROUGE-1: 0.1661 -> 0.2318 | Improved |
| add MCE-only loss (ablation) | SFT · Book Review | ROUGE-1: 0.0752 -> 0.1827 | Improved |
| add DCE token weighting only (ablation) | SFT · Book Review | ROUGE-1: 0.0752 -> 0.1958 | Improved |
| combine MCE-DCE alignment and token weighting (ablation) | DCE only · Book Review | ROUGE-1: 0.1958 -> 0.2318 | Improved |
| replace preference-driven tokens with random tokens (ablation) | NextQuill · Book Review | ROUGE-1: 0.2308 -> 0.1811 | Hurt |
| scale backbone from 3B to 7B | NextQuill · Book Review | ROUGE-1: 0.2318 -> 0.4116 | Improved |
| add NextQuill on Topic Writing | OPPU · Topic Writing | BERTScore: 0.7990 vs 0.7957 | No clear effect |
| plain SFT without history | Qwen base · Topic Writing | ROUGE-1: 0.0684 -> 0.0345 | Hurt |
| stratify by item popularity (ablation) | NextQuill vs ContextSFT · Book Review low-pop | ROUGE-1: 0.1651 -> 0.2079 | Improved |
| set alpha to 0.05 (ablation) | NextQuill · CD Review | multiple metrics | Improved |
Later papers that compared with this method
1 later accepted paper reports NextQuill as a baseline in their results tables; 1 of them report a better number on most of the results they share with it.
- Think-While-Generating: On-the-Fly Reasoning for Personalized Long-Form Generation (ICLR 2026 · better on 3 of 3 shared results)
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Aligning to Thousands of Preferences via System Message Generalization (NeurIPS 2024 · similarity 0.91)
- PersonalLLM: Tailoring LLMs to Individual Preferences (ICLR 2025 · similarity 0.91)
- Rethinking Personalization in Large Language Models at the Token Level (ICML 2026 · similarity 0.90)
- Spread Preference Annotation: Direct Preference Judgment for Efficient LLM Alignment (ICLR 2025 · similarity 0.90)
- HYDRA: Model Factorization Framework for Black-Box LLM Personalization (NeurIPS 2024 · similarity 0.89)
- PAD: Personalized Alignment of LLMs at Decoding-time (ICLR 2025 · similarity 0.89)
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.
