비주얼 노벨 창작자 유형 분류 및 LLM 기반 적응형 내러티브 분기 생성 파이프라인 연구

초록

In AI-assisted visual novel authoring environments, users perceive and interact with AI in fundamentally different ways. This study analyzes survey data collected from VN-AI Studio users (N=145) to classify creators into four types — Co-Creator, Director, Craftsman, and Learner — based on their AI role perception. These results confirm statistically significant differences in creator-type distribution across job categories (χ2=35.38, p<.001). This study proposes an adaptive narrative branch generation pipeline that maps the characteristics of each creator type to large language model (LLM) prompt parameters. A proof-of-concept (PoC) experiment demonstrates that semantically differentiated narrative branches can be generated from the same prologue when different type parameters are applied. The differentiation was further validated through cosine similarity analysis (mean similarity = 0.31). The proposed framework introduces a user-type-aware personalized narrative generation pipeline and offers a novel design perspective for AI-driven game authoring research. Furthermore, the study establishes both a theoretical and empirical foundation for future extensions toward multi-agent-based dynamic personalization systems.

키워드