AI Paper Atlas

Research map / Large language models / Alignment and preferences

CAReDiO: Enhancing Cultural Alignment of LLM via Representativeness and Distinctiveness Guided Data Optimization

Jing Yao, Xiaoyuan Yi, Jindong Wang, Zhicheng Dou, Xing Xie

ICML 2026 · Large language models › Alignment and preferences · approach: Steering and correcting LLM behavior

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 changedWhereMeasuredEffect
add representativeness objective only vs raw data scoring (Only Rep) (ablation)CAReDiO (Qwen2.5-7B-Instruct) · avg of CB-Easy, CB-Hard, Prism, GlobalOpinionQAaverage: 58.05 -> 59.43Improved
add distinctiveness objective to representativeness-only (ablation)CAReDiO (Qwen2.5-7B-Instruct) · avg of four benchmarksaverage: 59.43 -> 61.83Improved
add representativeness objective to distinctiveness-only (ablation)CAReDiO (Qwen2.5-7B-Instruct) · avg of four benchmarksaverage: 58.50 -> 61.83Improved
replace role-play with CAReDiO data fine-tuningQwen2.5-7B-Instruct · Prismrating (1-5): 3.364 -> 3.871Improved
fine-tune on CultureInstruction dataGemma-3-27B-IT · CulturalBench-Hardaccuracy: 46.59 -> 18.19Hurt
fine-tune on CultureBank dataQwen2.5-7B-Instruct · CulturalBench-Hardaccuracy: 38.90 -> 27.34Hurt
use CAReDiO instead of CultureLLM (WVS-augmented data)Qwen2.5-7B-Instruct · GlobalOpinionQA / WVSaccuracy: 57.11 vs 56.23 (GlobalOpinionQA)Hurt
fine-tune with CAReDiO data on GPT-4.1GPT-4.1 · CulturalBench-Hardaccuracy: 59.45 -> 63.54Improved
increase number of consensus individuals M (ablation)CAReDiO · average over 15 culturesperformanceImproved
increase optimization rounds (ablation)CAReDiO · average over 15 culturesperformanceImproved
increase training samples from 100 to 2,000 (ablation)CAReDiO · average over 15 culturesperformanceImproved
aggregate multi-role consensus instead of single role (ablation)GPT-5 · GlobalOpinionQAaccuracy: 58.63 -> 59.92Improved

Closest papers by meaning

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.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy