The topic evolution was investigated for network public opinion analysis. The properties of network public opinion information were analyzed firstly. Based on the properties, the latent semantics of textual data for network public opinion was described by using the topic model, and the text streams are modeled with a consideration of time for online analysis. Furthermore, a topic evolution method based on OLDA was proposed by incorporating the correlation of topics among time slices. The proposed method was experimentally verified to be efficient for detecting topic evolution of network public opinion. 