Academic

BiT-MCTS: A Theme-based Bidirectional MCTS Approach to Chinese Fiction Generation

arXiv:2603.14410v1 Announce Type: new Abstract: Generating long-form linear fiction from open-ended themes remains a major challenge for large language models, which frequently fail to guarantee global structure and narrative diversity when using premise-based or linear outlining approaches. We present BiT-MCTS, a theme-driven framework that operationalizes a "climax-first, bidirectional expansion" strategy motivated by Freytag's Pyramid. Given a theme, our method extracts a core dramatic conflict and generates an explicit climax, then employs a bidirectional Monte Carlo Tree Search (MCTS) to expand the plot backward (rising action, exposition) and forward (falling action, resolution) to produce a structured outline. A final generation stage realizes a complete narrative from the refined outline. We construct a Chinese theme corpus for evaluation and conduct extensive experiments across three contemporary LLM backbones. Results show that BiT-MCTS improves narrative coherence, plot str

Z
Zhaoyi Li, Xu Zhang, Xiaojun Wan · · 1 min read · 7 views
#cs.CL

arXiv:2603.14410v1 Announce Type: new Abstract: Generating long-form linear fiction from open-ended themes remains a major challenge for large language models, which frequently fail to guarantee global structure and narrative diversity when using premise-based or linear outlining approaches. We present BiT-MCTS, a theme-driven framework that operationalizes a "climax-first, bidirectional expansion" strategy motivated by Freytag's Pyramid. Given a theme, our method extracts a core dramatic conflict and generates an explicit climax, then employs a bidirectional Monte Carlo Tree Search (MCTS) to expand the plot backward (rising action, exposition) and forward (falling action, resolution) to produce a structured outline. A final generation stage realizes a complete narrative from the refined outline. We construct a Chinese theme corpus for evaluation and conduct extensive experiments across three contemporary LLM backbones. Results show that BiT-MCTS improves narrative coherence, plot structure, and thematic depth relative to strong baselines, while enabling substantially longer, more coherent stories according to automatic metrics and human judgments.

Executive Summary

This article presents BiT-MCTS, a theme-driven framework for generating long-form linear Chinese fiction. By employing a 'climax-first, bidirectional expansion' strategy motivated by Freytag's Pyramid, BiT-MCTS operationalizes a novel approach to narrative generation. The method extracts a core dramatic conflict, generates an explicit climax, and utilizes a bidirectional Monte Carlo Tree Search to expand the plot. Experimental results demonstrate that BiT-MCTS improves narrative coherence, plot structure, and thematic depth relative to strong baselines. The framework enables the generation of longer, more coherent stories according to automatic metrics and human judgments. The study's findings have significant implications for the development of large language models and the creation of engaging, structurally sound narratives.

Key Points

  • ▸ BiT-MCTS operationalizes a novel 'climax-first, bidirectional expansion' strategy for narrative generation.
  • ▸ The framework employs a bidirectional Monte Carlo Tree Search to expand the plot.
  • ▸ Experimental results demonstrate improvements in narrative coherence, plot structure, and thematic depth.

Merits

Strength in Operationalization

The 'climax-first, bidirectional expansion' strategy provides a structured approach to narrative generation, enabling the creation of longer, more coherent stories.

Advancements in Large Language Models

BiT-MCTS demonstrates the potential for large language models to generate engaging, structurally sound narratives, expanding their capabilities beyond premise-based or linear outlining approaches.

Demerits

Limited Scope

The study focuses on generating Chinese fiction, limiting the framework's applicability to other languages and narrative genres.

Dependence on Monte Carlo Tree Search

The framework's reliance on a bidirectional Monte Carlo Tree Search may introduce computational complexity and limitations in scalability.

Expert Commentary

The BiT-MCTS framework presents a significant advancement in narrative generation, demonstrating the potential for large language models to produce engaging, structurally sound stories. However, the study's limitations, such as the focus on Chinese fiction and reliance on Monte Carlo Tree Search, highlight areas for future research and development. The findings have important implications for the development of creative writing tools, educational settings, and the entertainment industry, underscoring the need for a nuanced understanding of the role of creativity in artificial intelligence.

Recommendations

  • ✓ Further research into the adaptability of BiT-MCTS to other languages and narrative genres.
  • ✓ Investigation into the potential applications of the framework in other creative writing contexts, such as poetry or screenwriting.

Sources

Related Articles