As microblog services become increasingly popular, spatial-temporal text data has increased explosively. Many studies have proposed methods to spatially and temporally analyze an event, indicated by the text data. These studies have aimed a extracting the period and the location in which a specified topic frequently occurs. In this paper, we focus on a system that detects hot topic in a local area and during a particular period. There can be a variation in the words used even though the posts are essentially about the same hot topic. We propose a classification method that mitigates the variation of posted words related to the same topic.