Social media is a highly influential platform for sharing messages, photos, and videos. Understanding public perception through its vast data stream is essential. This study introduces a two-stage clustering method to extract coarsegrained topics from social media text data. First, graph clustering extracts micro-clusters from graphs generated based on the similarity of user posts, with each micro-cluster representing a fine-grained topic. The time series of these microclusters are then analyzed in the second stage through time series clustering to reveal coarse-grained topics. In this study, we consider applying this method to Yahoo! Japan News Comments related to the election of two specific candidates in Japan. This is expected to extract people's reactions to the candidates before and after the election.
