This paper reports a social media analysis studyproposed by the Beijing Municipal Institute of Urban Planningand Design. The purpose is to explore techniques that can help theurban planning administrations to improve the social sensing andsocial perception abilities under the evolving data and technologyenvironments. A framework integrating a comprehensive set oftext mining algorithms is presented to conduct topic modeling, text clustering, event evolution detection, sentiment analysis, opinion mining, and information extraction on user-generatedcontents in Chinese social media. A domain ontology of Beijingurban planning is constructed to facilitate the text mining processes. Evaluations on two large, real-world datasets composed ofmicroblogs and WeChat articles about the residential communityand school education in Beijing demonstrate the effectiveness ofour framework. The study illustrates the power of combiningmachine learning with knowledge-based, semantic approaches inanalyzing social media for the domain of interest.