Research map / Large language models / Alignment and preferences
StoryAlign: Evaluating and Training Reward Models for Story Generation
Haotian Xia, Hao Peng, Yunjia Qi, Bin Xu, Lei Hou, Juanzi Li
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 Premise Back-Generation training pairs (ablation) | StoryReward-Qwen · StoryRMB | Average accuracy: 73.9 -> 75.0 | Improved |
| add Human-Guided Continuation training pairs (ablation) | StoryReward-Qwen · StoryRMB | Average accuracy: 72.8 -> 75.0 | Improved |
| add Prompt-Guided Rewriting training pairs (ablation) | StoryReward-Qwen · StoryRMB | Average accuracy: 69.9 -> 75.0 | Improved |
| add training on human-written stories and preference pairs (StoryReward-Qwen vs Qwen3-8B base) | Qwen3-8B · StoryRMB | Average accuracy: 53.5 -> 75.0 | Improved |
| train StoryReward on Llama-3.1-8B-Instruct | Llama3.1-8B-Instruct · StoryRMB | Average accuracy: 24.3 -> 61.4 | Improved |
| use exact human-written stories as chosen instead of human-guided LLM continuation (ablation) | StoryReward training data · StoryRMB | Average accuracy: below 30% | Hurt |
| ablated components drop on coherence/relevance (ablation) | StoryReward-Qwen · StoryRMB Coherence | accuracy: 77.5 -> 82.3 | Mixed |
| evaluate reward models on Human-LLM pairs instead of LLM-LLM pairs | existing reward models · StoryRMB subsets | accuracy | Hurt |
| select best of 16 stories using StoryReward vs other reward models | Llama-3.1-70B-Instruct generations (MoPS) · BoN head-to-head | inferior-story selection rate: fewer than 30% | Improved |
Closest papers by meaning
- Agents' Room: Narrative Generation through Multi-step Collaboration (ICLR 2025 · similarity 0.86)
- StoryGPT-V: Large Language Models as Consistent Story Visualizers (CVPR 2025 · similarity 0.85)
- ViStoryBench: Comprehensive Benchmark Suite for Story Visualization (CVPR 2026 · similarity 0.85)
- ReMoDetect: Reward Models Recognize Aligned LLM's Generations (NeurIPS 2024 · similarity 0.85)
- Artificial Hivemind: The Open-Ended Homogeneity of Language Models (and Beyond) (NeurIPS 2025 · similarity 0.85)
- One-Prompt-One-Story: Free-Lunch Consistent Text-to-Image Generation Using a Single Prompt (ICLR 2025 · similarity 0.85)
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.
