LangChain과 속성 기반 감성 분석 모델을 활용한 시스템 개발에 대한 연구: 부동산 도메인을 중심으로

This paper proposes a question-answering system for generating user-customized responses in the real estate domain by leveraging aspect-based sentiment analysis and the LangChain framework. In the real estate market, actual residence reviews related to apartments provide crucial information to potential buyers. However, the vast amount of text and the diverse aspects covered in each review present limitations in efficiently analyzing and understanding this information. To address this issue, we trained a model to analyze sentiments of apartment review data based on residential values that people focus on. Additionally, by combining the sentiment analysis model with Retrieval-Augmented Generation (RAG) technology, we developed a system that extracts information from various sources and generates reliable answers to queries. For this purpose, we constructed an apartment review dataset for aspect-based sentiment analysis and integrated the trained model with LangChain and a vector database to build the system. The proposed system not only presents actual residence reviews of apartment complexes by residential values but also provides helpful information by linking with external data.