Abstract: Information tracking is to detect the information related to a certain emergency from the text stream. The dynamic evolvement characteristic is one of the important reasons which lead to the high miss rate, and we propose an emergency information tracking method based on the multi-classification to deal with this problem, which regard the multi focuses appearing during the event evolvement as the multi classes. In this tracking method, we propose and explore a post representation model based on the context and latent semantic analysis and an event representation model based on the multi-focus. We apply the proposed information tracking method in the earthquake information, and the experimental results show that the proposed method can improve the performance of the earthquake information tracking system.