Planning Beyond Text: Graph-based Reasoning for Complex Narrative Generation

2026-01-04 · Source: cs.CL updates on arXiv.org · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Emerging Technologies & Innovation · Depth: Expert, extended

Summary

PLOTTER is a novel framework designed to enhance complex narrative generation by employing graph-based representations for planning, rather than direct sequential text. It addresses common issues in large language model (LLM) narrative generation, such as maintaining global coherence, logical consistency, and character development, which often result in monotonous scripts with structural flaws. PLOTTER utilizes an "Evaluate-Plan-Revise" cycle on both an event graph, capturing causal structure, and a character graph, modeling inter-character relationships. This process diagnoses and repairs topological issues under rigorous logical constraints, optimizing the narrative skeleton before full context generation. Experiments across diverse narrative scenarios, using LLM backbones like GPT-4.1, DeepSeek-R1, and Qwen3, demonstrate that PLOTTER significantly outperforms baselines in narrative coherence, thematic engagement, and character development, with a default cost of 1.68 USD per script for K=3 iterations.

Key takeaway

Research scientists developing narrative AI should consider adopting graph-based planning frameworks like PLOTTER to overcome limitations of text-centric generation. By explicitly modeling causal and character relationships through event and character graphs, you can achieve superior narrative coherence, deeper characterization, and more engaging plots, even with a controllable computational cost of 1.68 USD per script for high-quality output. This approach fundamentally changes how LLMs reason about long-form narratives, moving beyond sequential text to structural logic.

Key insights

Graph-based narrative planning significantly enhances LLM-generated story coherence and character development over text-based methods.

Principles

• Narrative planning on structural graphs improves long-context reasoning.
• Iterative graph refinement resolves inconsistencies at a causal level.
• Multi-agent critique and constrained editing ensure structural integrity.

Method

PLOTTER constructs and refines event and character graphs using an Evaluate-Plan-Revise cycle with specialized agents and a constrained editor, then serializes these graphs for state-aware script synthesis.

In practice

• Use graph representations for complex story planning.
• Implement multi-agent critique for structural diagnosis.
• Apply atomic graph-editing operations for refinement.

Topics

• PLOTTER Framework
• Graph-based Narrative Planning
• Large Language Models
• Event Graph
• Character Graph

Best for: Research Scientist, AI Scientist, Machine Learning Engineer, NLP Engineer

Related on AIssential

• GraphLit: Learning Text-Enriched Dynamic Character Network Representations for Literary Study — GraphLit integrates dynamic character networks with textual context via self-supervised learning to enhance literary analysis.
• LLM Friendly Graphs for AI Memory — From Triples to Facts — Larger, context-rich labels and subgraph representations enhance LLM comprehension of complex facts in knowledge graphs.
• GraphScout: Empowering Large Language Models with Intrinsic Exploration Ability for Agentic Graph Reasoning — GraphScout enables LLMs to autonomously explore knowledge graphs, internalizing reasoning through synthesized training data.
• GraphDC: A Divide-and-Conquer Multi-Agent System for Scalable Graph Algorithm Reasoning — GraphDC uses a divide-and-conquer multi-agent system to scale LLM-based graph reasoning by decomposing complex graphs.
• StoryCoder: Narrative Reformulation for Structured Reasoning in LLM Code Generation — Narrative reformulation of code problems significantly improves LLM code generation by structuring reasoning.
• LLMs+Graphs: Toward Graph-Native, Synergistic AI Systems — Integrating LLMs with graph computation creates synergistic AI systems that overcome LLM reasoning limitations.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by cs.CL updates on arXiv.org.