Research map / Large language models / Language, safety and interpretability
Evaluating Text Creativity across Diverse Domains: a Dataset and Large Language Model Evaluator
Qian Cao, Xiting Wang, Yuzhuo Yuan, Yahui Liu, Fang Luo, Ruihua Song
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 |
|---|---|---|---|
| train on CreataSet pairwise data | Qwen2.5-7B-Instruct · CreataSet test | agreement rate: 0.574 -> 0.745 | Improved |
| train on CreataSet pairwise data | Qwen2.5-14B-Instruct · CreataSet test | agreement rate: 0.661 -> 0.762 | Improved |
| train on CreataSet pairwise data | Qwen2.5-14B-Instruct · CreataSet test, Lyrics | F1: -2.1% | Hurt |
| provide a shared instruction as context for annotators | human annotation · 400 text pairs | ICC: 0.59 -> 0.75 | Improved |
| add swapped-position augmentation in training (ablation) | CrEval-7B · CreataSet test | consistency | Improved |
| add human original responses to training data (ablation) | CrEval · CreataSet test | agreement | Improved |
| add synthetic responses to training data (ablation) | CrEval · CreataSet test | agreement | Improved |
| add negative sampling of unrelated responses (ablation) | CrEval · CreataSet test | agreement | Improved |
| increase training data beyond 100K samples (ablation) | CrEval · CreataSet test | F1/Kappa/agreement | No clear effect |
| use classification loss instead of Bradley-Terry loss (ablation) | CrEval-7B · CreataSet test | consistency: 0.846 -> 0.920 | Improved |
| use Qwen2.5-14B instead of 7B backbone (ablation) | CrEval · CreataSet test | agreement rate: 0.745 -> 0.762 | Improved |
| replace contextual DPO rejects with random negative rejects (ablation) | DPO on Qwen2.5-7B-Instruct · win rate vs GPT-4o-mini | win rate | Hurt |
Closest papers by meaning
- CreBench: Human-Aligned Creativity Evaluation from Idea to Process to Product (AAAI 2026 · similarity 0.92)
- AI as Humanity’s Salieri: Quantifying Linguistic Creativity of Language Models via Systematic Attribution of Machine Text against Web Text (ICLR 2025 · similarity 0.92)
- IDEAFix: Evaluation Framework for Creative Defixation Prompting in LLMs (NeurIPS 2026 · similarity 0.88)
- CLAWS:Creativity detection for LLM-generated solutions using Attention Window of Sections (NeurIPS 2025 · similarity 0.88)
- Quality-Diversity through AI Feedback (ICLR 2024 · similarity 0.88)
- Within-Model vs Between-Prompt Variability in Large Language Models for Creative Tasks (NeurIPS 2026 · similarity 0.87)
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.
