Kanami Sugiyama, Tetsuji Kuboyama, Hirofumi Miwa, Takeaki Uno
2022 289-294, Dec 2, 2022 Peer-reviewed
A text analysis was applied to the election manifestos. Two kinds of clustering methods were utilized and compared: microclustering that extracted dense substructures from the network constructed based on the similarity of the documents, and the LDA model, which is a kind of topic model. The microclustering gave many clusters that were easier to understand their topics, especially for the party of candidates. Furthermore, regression analysis was applied to the obtained clusters, and the tendency of candidates with personal-oriented was elucidated. The results correspond to previous studies, such as the change of the electoral rule and the characteristics of each political party, without any manual topic interpretation process as in the previous studies.