프롬프트 엔지니어링 기반 ChatGPT의 토픽 모델링 자동화 구현 연구

This study proposes a methodology that can perform topic modeling analysis using only natural language-based instructions, utilizing ChatGPT, an interactive AI model, and Prompt Engineering techniques. Existing topic modeling has limitations in that it requires a high level of programming understanding and complex model setting by utilizing statistical-based algorithms such as LSA, LDA, and NMF. However, this study gave GPT expertise by defining a structured topic modeling analysis procedure based on natural language and assigning a ""data scientist"" role using prompt engineering techniques. As a result, we experimentally presented an approach to automating the entire process, including topic number determination, model selection, visualization, and consistent analysis, for real news data according to domain expertise and context-based interpretation, which were difficult to implement in existing automation models. The automatically generated Python code during the analysis process was re-executed in a separate development environment to verify the accuracy and reliability of the GPT generation code. In addition, the GPT Hallucination phenomenon was prevented by fetching verified external document information through ChatGPT's [knowledge] function and securing the objectivity of analysis. The GPT model of this study had some restrictions on the ChatGPT operating environment such as Korean morphological analysis, deep learning-based topic modeling such as BERTopic, and OpenAPI linkage. However, it is significant as a basic study for the possibility of automating ChatGPT prompt-based data analysis and expanding the practical use of ChatGPT.