A text-based mining approach for real estate policy impact monitoring and analysis

Text mining has been widely used in economic analysis. Besides extracting individual opinion from a single text, it is more important to accurately aggregate group opinions from mass text files. This research proposed a text mining framework for constructing policy impact indicators to monitoring the public opinions on real estate market policies. A real estate domain dictionary has been developed based on the released policy announcements in Chinese real estate government websites. And corresponding public opinion index are constructed to evaluate the impact of real estate policies on the target market. In addition, we have improved the handling of emotional words in the public opinion dictionary. Word2Vec and So-PMI models are used to identify the polarity of trend words, and the score of each sentiment words is calculated based on its co-occurrences with the trends words using a modified label propagation algorithm. Beijing real estate market is taken as the test bed. Empirical study show that the policy public opinion index explains a significant part of Beijing housing price changes. In-sample and out-of-sample forecasts of Beijing housing prices indicate that the policy public opinion index is capable of increase the prediction accuracy.