Abstract: An in-depth analysis of fake house microblog commentary based on text mining technology, a text feature analysis by means of text feature extraction, word frequency analysis and feature association analysis, and an analysis of sentiment tendency, may help provide an insight into the deeper meaning of the commentary, the real concerns and sentimental attitude of the public, and the formation mechanism underlying the concerns and attitude. Such an insight enables the public to optimize their behavior strategies, intermediary enterprises to improve their services and authorities to change their supervision mode. The study shows that:(1) public sentiment towards fake house ads and the overall housing market is predominantly negative, as evidenced by 72.1% of the public holding a negative attitude and 69.3% holding a highly negative attitude; (2) there is a misalignment in the focus of public attention on the four levels of housing quality, service level, individual perception and supervision; (3) the core factors for whether the public sentiment is positive or negative lie in the quality of available houses and the level of intermediary service. The intensity of regulation plays a smaller role in generating positive sentiment than in generating negative sentiment; and (4) the choice of the business model of intermediary enterprises affects the degree of importance they attach to fake house ads, which indirectly affects the change in public sentiment. The above findings have implications for revealing the nature of fake house ads, digging deeper into the public's real inner needs and promoting the benign development of the real estate market.

Key words: housing information, text mining, sentiment feature analysis