Abstract: The in-depth exploration of the text data contained in social media facilitates efficient analysis of time and space. This paper proposes a new social media topic mining method based on the concept of co-word network and community detection. The method uses term frequency-inverse document frequency (TF-IDF) analysis to identify the key words of the messages automatically. Based on the problem whether the microblogs contain the same key words or not, we put forward the concept of microblog co-word network with microblog as the node. The network combined with the Louvain community detection algorithm is used to classify the microblogs into different clusters with topics. The proposed method is an unsupervised method. The advantage of this method is that there is no need to specify the number of clusters. Experiments demonstrate that the performance of the proposed method is better than the commonly used latent dirichlet allocation (LDA) model on both precision and recall. Taking the collected microblogs during the 2012 Beijing rainstorm as the case study, the method is used to conduct in-depth mining and time-space analysis of the microblogs dataset. The results demonstrate that the proposed method is effective in real world applications.