Classification of Saikaku’s works using topic modelling

Abstract This study quantitatively analysed the thematic structure and temporal transitions in Saikaku Ihara’s ukiyo-zōshi using latent Dirichlet allocation (LDA), dynamic topic modelling (DTM), and Word2Vec. Previous studies have discussed the themes and genres of Saikaku’s works; however, comprehensive quantitative analysis has been lacking. Applying LDA to twenty-four works, this study examined topic transitions. The results confirmed previous research findings, showing that early works primarily focused on Kōshokumono (romantic novel), while later works covered broader themes. Additionally, LDA identified distinct topics within each chapter, subdividing Kōshokumono into emotions and love (Topic 3), pleasure quarters (Topic 4), and personal romantic history (Topic 5). A comparison with Munemasa’s classification quantitatively demonstrated the relationship between topics and literary genres. DTM confirmed that early works centred on romance, while later works incorporated Bukemono (samurai story) and Chōninmono (merchant story), reinforcing the LDA findings. Word2Vec revealed semantic shifts in key terms: fukashi (deep and intense) transitioned from intellectual and emotional depth to economic value, namida (tears) shifted from personal grief to collective sentiment and economic metaphors, and toru (take, pick, and capture) and kane (money and chime) evolved from denoting wealth acquisition to social relationships and ethical value judgements. By integrating LDA, DTM, and Word2Vec, this study’s quantitative examination of thematic transitions in Saikaku’s works provides new insights into the structural and lexical evolution of ukiyo-zōshi beyond conventional classifications.