기계학습 기술을 이용한 부동산 감성지수 개발 모형 연구

The market sentiment analysis is a useful way to understand the participants’ sentiment in the real estate market by extracting their opinion, attitude and tendency from the irregular text data. This study aims to develop the sentiment index model, using real estate-related online newspaper articles. The procedures are as follows: First, online news articles are web-crawled from major daily and economic news websites. Second, topics and words are extracted through the Latent Dirichlet Allocation (LDA) topic analysis. Third, the sentiment dictionary is established using the TextRank algorithm with sentences which contain the words extracted by the topic analysis in the second step. Finally, using the Term Frequency - Inverse Document Frequency (TF-IDF) and the Naive Bayes classification model, we assign polarity to the sentences and calculate the weights, producing the monthly real estate sentiment index. The result shows a good performance of the proposed method with an accuracy of 88%. The model is more advanced than existing methods which have been used in the real estate studies, proposing a novel analytical framework and model for the unstructured big data analysis. It also provides an index which reflects participants‘ sentiment immediately and flexibly, therefore helps to explain and predict changes in the real estate market.