This paper proposes a method for analyzing time series topic transition based on micro-clusters to present different situations that show people's reactions to topical problems on the Web. To form micro-clusters, we leverage our original data polishing algorithm developed by one of the authors. Our method shows that micro-clusters efficiently represent the dynamics of topic transitions: for example, events cause sudden changes in the number of clusters. This implies that there were increases or decrease of diversity of cluster contents that correspond to people's feelings and opinions to the topic. To show the method's effectiveness, we conducted an experiment on tweets targeting rumors of a petrochemical complex explosion just after the Great East Japan Earthquake in 2011. Our method easily identifies the following phases in topic transitions. First, people post the real story. Second, rumors circulate about the explosion. Finally, the rumors were corrected by the government and gradually disappeared.
