RAG-LLM(Retrieval Augmented Generation-Large Language Model) 기반 부동산 뉴스 분석을 통한 아파트 매매가격지수 예측 모델 개발

This study explores the effectiveness of retrieval augmented generation (RAG) with large language models (LLMs) in processing unstructured data in real estate. By connecting pretrained LLMs with external knowledge sources, the RAG methodology reduces hallucinations while generating reliable domain-specific responses. We construct an RAG-LLM model using real estate news articles to predict directional movements in Seoul apartment price indices. Analysis spanning 2015‒2024 reveals that RAG-LLM predictions serve as significant leading indicators for apartment price changes, achieving 85% accuracy and 93% area under the curve - receiver operating characteristic (AUC-ROC). These RAG-LLM predictions outperform those based on the autoregressive integrated moving average (ARIMA) model by more than 11%. Employing chain-of-thought prompts with RAG further enhances performance, with improved metrics observed when evaluated using ARIMA with exogenous variables. Our findings suggest that the RAG-LLM approach represents a significant advancement in real estate analysis by effectively leveraging unstructured data and overcoming the limitations of traditional sentiment analysis through complex contextual understanding. The RAG-LLM approach also enables near real-time forecasting, significantly boosting its practical utility in real-world applications.