AI Paper Atlas

Research map / Large language models / Alignment and preferences

Modeling Attributional Style at Scale: A Dataset and Analysis for Psychological Attribution Assessment and Reframing

Qiang Zhou, Hanzhen Zhu, Pan Wang, Rui Tu, Huaizhi Qu, Zhuoran Wang, Xin Hu, Lei Li, Tianlong Chen, Jingtong Hu

ICML 2026 · Large language models › Alignment and preferences · approach: Robust and calibrated DPO

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 supervised fine-tuning on ASTD (LoRA) versus 3-shot prompting at same parameter countGemma 3-1B · ASTD classificationaverage accuracy (%): 33.44 -> 92.66Improved
add 3 in-context examples to Llama 3.3-70BLlama 3.3-70B · ASTD GAS-SPASaccuracy (%): 71.09 -> 88.26Improved
add 3 in-context examples to small Gemma 3-1BGemma 3-1B · ASTD classificationaverage accuracy (%): 39.79 -> 33.44Hurt
scale LLM size (prompt-only)Gemma 3 family · ASTD classification, 3-shotaverage accuracy (%): 33.44 (1B) -> 45.54 (4B) -> 79.90 (12B)Improved
train LoRA only on GAS-SPAS then test on IAS-EAS (cross-dimension transfer) (ablation)LoRA-tuned LLM · ASTD IAS-EASaccuracy (%): 95.33 -> 79.42Hurt
replace TF-IDF + LR with neural fine-tuned modelsattribution classifier · ASTD classificationaverage accuracy (%): 91.33 -> 97.33Improved
add DPO on LLM-evaluator preference pairsDeepSeek-R1-8B and Gemma 3-4B reframing · ASTD reframing (GAS2SPAS)constructive coping (1-5): 3.823 -> 4.715Improved
add DPOreframing LLMs · ASTD reframing, SAS2UASconstructive coping (1-5): 3.004 -> 4.137Improved
add DPOreframing LLMs · ASTD reframing (all tasks)coherence (1-5): GAS2SPAS 4.964 -> 4.990No clear effect
use SFT on preferred outputs instead of DPO (ablation)reframing LLMs · ASTD reframingcoherence: SFT -0.19 vs DPO +0.08Hurt
use DPO instead of SFT-on-y+ (ablation)reframing LLMs · ASTD reframinggains on the other dimensions: 3-5x larger gainsImproved
use reasoning-optimized model at smaller sizeprompt-only LLMs · ASTD classificationaverage accuracy (%): DeepSeek-R1-8B 80.26 vs Gemma 3-12B 79.90 (3-shot)Improved

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