User generated contents in social media platforms provide important and timely indicators on the spontaneous and often genuine views of users, fans, and customers on a wide range of topics. It is thus invaluable to obtain actionable insights from such live streaming contents. In this talk, I consider the problem of topic-specific micro-post representation in which we aim to develop accurate language models that encode micro-posts with arbitrary lengths into vectors of fixed length such that semantically similar micro-posts are close in the vector space. I will show the application of our model in Brandtology that is defined as the science of studying brands, customers, and the interaction between the two in the context of social networks. In particular, I will talk about two issues in Brandtology, namely churn prediction and emerging topic detection for brands in social media.