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
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 supervised fine-tuning on ASTD (LoRA) versus 3-shot prompting at same parameter count | Gemma 3-1B · ASTD classification | average accuracy (%): 33.44 -> 92.66 | Improved |
| add 3 in-context examples to Llama 3.3-70B | Llama 3.3-70B · ASTD GAS-SPAS | accuracy (%): 71.09 -> 88.26 | Improved |
| add 3 in-context examples to small Gemma 3-1B | Gemma 3-1B · ASTD classification | average accuracy (%): 39.79 -> 33.44 | Hurt |
| scale LLM size (prompt-only) | Gemma 3 family · ASTD classification, 3-shot | average 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-EAS | accuracy (%): 95.33 -> 79.42 | Hurt |
| replace TF-IDF + LR with neural fine-tuned models | attribution classifier · ASTD classification | average accuracy (%): 91.33 -> 97.33 | Improved |
| add DPO on LLM-evaluator preference pairs | DeepSeek-R1-8B and Gemma 3-4B reframing · ASTD reframing (GAS2SPAS) | constructive coping (1-5): 3.823 -> 4.715 | Improved |
| add DPO | reframing LLMs · ASTD reframing, SAS2UAS | constructive coping (1-5): 3.004 -> 4.137 | Improved |
| add DPO | reframing LLMs · ASTD reframing (all tasks) | coherence (1-5): GAS2SPAS 4.964 -> 4.990 | No clear effect |
| use SFT on preferred outputs instead of DPO (ablation) | reframing LLMs · ASTD reframing | coherence: SFT -0.19 vs DPO +0.08 | Hurt |
| use DPO instead of SFT-on-y+ (ablation) | reframing LLMs · ASTD reframing | gains on the other dimensions: 3-5x larger gains | Improved |
| use reasoning-optimized model at smaller size | prompt-only LLMs · ASTD classification | average accuracy (%): DeepSeek-R1-8B 80.26 vs Gemma 3-12B 79.90 (3-shot) | Improved |
Closest papers by meaning
- CAREBench: Evaluating LLMs' Emotion Understanding by Assessing Cognitive Appraisal Reasoning (NeurIPS 2026 · similarity 0.84)
- Thousand Voices of Trauma: A Large-Scale Synthetic Dataset for Modeling Prolonged Exposure Therapy Conversations (NeurIPS 2025 · similarity 0.82)
- Mental-Perceiver: Audio-Textual Multi-Modal Learning for Estimating Mental Disorders (AAAI 2025 · similarity 0.81)
- Diverging Preferences: When do Annotators Disagree and do Models Know? (ICML 2025 · similarity 0.81)
- Rethinking Psychometric Evaluation of LLMs: When and Why Self-Reports Predict Behavior (NeurIPS 2026 · similarity 0.81)
- Unveiling the Cognitive Compass: Theory-of-Mind–Guided Multimodal Emotion Reasoning (ICLR 2026 · similarity 0.81)
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.
