用知识图谱增强大模型叙事，让用户可编辑图谱来控制长篇故事生成。
Guiding Generative Storytelling with Knowledge Graphs
- 构建双阶段流程：先用知识图谱辅助生成，再允许用户编辑图谱调整故事。
- 用户在15人测试中发现，动作导向的故事结构更清晰，质量提升明显。
- 编辑知识图谱让使用者感觉有掌控感，过程有趣且互动性强。
大型语言模型在故事生成方面展现出巨大潜力，但在保持长篇连贯性及实现用户友好控制方面仍面临挑战。检索增强生成（RAG）已被证明能有效减少文本生成中的幻觉；尽管先前研究探索过基于知识图谱（KG）的叙事，但本工作聚焦于KG辅助的长篇生成，并在两阶段用户研究中引入可编辑的KG与LLM生成结合。研究探讨了知识图谱如何通过提升叙事质量和支持用户驱动修改来增强基于大模型的故事创作。我们提出一种KG辅助的故事生成流程，并在15名参与者的用户研究中进行评估。参与者创建提示、生成故事并编辑知识图谱以塑造叙事内容。定量与定性分析显示，在设定条件下，行动导向且结构明确的故事质量显著提升，但内省型故事未见改进。参与者报告称编辑知识图谱带来强烈控制感，形容该体验具有吸引力、交互性和趣味性。
原文摘要 · Abstract (English)
Large language models (LLMs) have shown great potential in story generation, but challenges remain in maintaining long-form coherence and effective, user-friendly control. Retrieval-augmented generation (RAG) has proven effective in reducing hallucinations in text generation; while knowledge-graph (KG)-driven storytelling has been explored in prior work, this work focuses on KG-assisted long-form generation and an editable KG coupled with LLM generation in a two-stage user study. This work investigates how KGs can enhance LLM-based storytelling by improving narrative quality and enabling user-driven modifications. We propose a KG-assisted storytelling pipeline and evaluate it in a user study with 15 participants. Participants created prompts, generated stories, and edited KGs to shape their narratives. Quantitative and qualitative analysis finds improvements concentrated in action-oriented, structurally explicit narratives under our settings, but not for introspective stories. Participants reported a strong sense of control when editing the KG, describing the experience as engaging, interactive, and playful.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。