검색증강생성(RAG) 기반 한국형 상표 생성 플랫폼 프로토타입 연구 - 생성형 인공지능 프롬프트 키워드 설계를 중심으로 -

This study aims to address the issue of cultural bias in generative artificial intelligence and to establish an automated trademark design system that reflects the unique visual language of Korea. By applying Retrieval-Augmented Generation (RAG), the research integrates the Korean trademark information database (KIPRIS) into the design process, thereby enabling culturally grounded outcomes. The research selectively extracted industry-specific datasets from the NICE classification where visual tendencies were more prominent. A structured prompt library was developed based on Midjourney outputs, categorized by attributes such as style, components, color and tone, textual inclusion, and emotional vocabulary. Subsequently, co-occurrence network analysis and centrality measures were applied to identify the structural relationships between prompt keywords. The analysis revealed that core keywords such as korean, logo, simple, white, and background consistently dominated the structure of prompt language, serving as central nodes in the network. Meanwhile, bridging keywords such as bowl, line, traditional, and pattern played a mediating role across clusters, connecting visual properties with cultural symbols. These findings demonstrate how cultural and aesthetic coherence can be preserved within generative AI design processes. The study proposes an alternative framework for securing cultural alignment in AI-driven trademark design by embedding Korean visual identity into generative systems. Furthermore, it suggests that future research incorporating user feedback and multimodal model expansion could enhance both the practicality and the ethical validity of design automation.