Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

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

Zhaoyi Li, Xu Zhang, Xiaojun Wan

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.

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

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.
Paper Structure (36 sections, 6 equations, 3 figures, 7 tables, 1 algorithm)

This paper contains 36 sections, 6 equations, 3 figures, 7 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Methodology
  3. Overview
  4. Conflict and Climax Generation
  5. Bidirectional MCTS
  6. Node representation.
  7. Evaluator (reward function).
  8. High-level procedure.
  9. Selection.
  10. Expansion.
  11. Simulation
  12. Backpropagation
  13. Outline Refinement
  14. Segmented Fiction Generation
  15. Experiments
  16. ...and 21 more sections

Figures (3)

  • Figure 1: Comparison of fiction outline generation methods: Sequential outline generation leads to overly formatic narratives, while bidirectional MCTS can generate diverse and creative outlines.
  • Figure 2: An overview of the four-stage fiction generation pipeline, which proceeds: (1) Conflict and Climax Generation establishes the core conflict and the core climax. (2) Bidirectional MCTS Exploration searches forward and backward from the climax for coherent and creative plot outlines. (3) Outline Refinement refines the rough outline generated by Bidirectional MCTS and segmented. (4) Segmented Fiction Generation expands the refined outline into the final, fluent narrative fiction. The pipeline transforms an abstract theme into a complete, structured long-form fiction. Note that the example English texts are translated from Chinese texts for better understanding.
  • Figure 3: Win rates of BiT-MCTS when generating stories of different lengths in pairwise comparisons across ten dimensions (DeepSeek-V3, short average length: 4669 words vs. long average length: 8059 words).