서사 구조를 따르는 LLM 기반 스토리 생성 연구

• 영문명: LLM-Based Story Generation Aligned with Narrative Structure
• 발행기관: 한국컴퓨터게임학회
• 저자명: 지은수(Eunsu Ji) 유태경(Tae-Kyung Yoo)
• 간행물 정보: 『한국컴퓨터게임학회논문지』제38권 3호, 61~72쪽, 전체 12쪽
• 주제분류: 공학 > 컴퓨터학
• 파일형태: PDF
• 발행일자: 2025.07.31

국문 초록

This study proposes a structured story generation system based on traditional narrative theory using a large language model (LLM). The proposed system sequentially constructs a three-act structure and a 15-step narrative structure of “Save the Cat!” based on the log line, character information, and genre input by the user, and each step is embodied through hierarchical generation and prompt chaining. In particular, when user modifications occur, the changes are designed to be automatically reflected in the upper and lower stages to maintain narrative consistency and logic. To verify the effectiveness of this system, human-written narratives, single-prompt-based generative narratives, and generative narratives from the proposed system were compared and evaluated, yielding excellent results in terms of narrative structure fidelity and logic. By demonstrating the structural controllability of LLM-based story generation, this study suggests applicability in the field of digital content production in the future.

목차

1. 서론

2. 관련 연구

3. 구조 설계 및 구현

4. 실험 설계

5. 결론

Acknowledgement

참고문헌

키워드

Large Language Models Story Generation Narrative Structure Prompt Chaining Hierarchical Story Generation