With the rapid development of the We-Media, monitoring and guidance of public opinion becomes a significant research subject. Traditional topic detection methods in microblog data analytics encounters the problems of high computational complexity, low real-time and recall rate. An improved algorithm based on emotions and word co-occurrence detection is proposed in this paper aiming at solving these problems. It builds a emotional subspace model through co-occurrence relation of sentiment words in hot events, and classifies the flow of information in weibos. Finally, it gets the aim of topic detection via extracting the subject in the corresponding category. The experimental results carries out on the microblog content corpus of NLPIR and verifies that this method can effectively detect news topic from the massive microblog information and realize the news topic tracking.