This paper addresses the outline of the historical development of music and dance that supports the unique parlors Keihan (Kyoto and Osaka) in Japan, and arrangement and classification of the song format and song content for the data of about 600-700 songs of ``Dance song'' and ``Performance song.''

The Keihan hanayanagi community has preserved and developed the world of ``Mai'' (dance) along with performances of ``jiuta'' and ``kamigata uta,'' as well as ``hauta'' and ``kouta.'' ``Jiuta'' has the aspect of traditional vocal music, while ``Kamigata Uta'' is light and cheerful shamisen music from the urban areas of Keihan.

The purpose of this study is to analyze the characteristics of the ``song words'' of the <dance songs> and <performance songs> of the “jiuta/kamigata songs” of Keihan by text mining. First, interviews were conducted with persons concerned and materials were collected to organize the content of the songs that are unique to the Kyoto-Osaka zashiki, and from the collected materials 600-700 songs were organized, classified, and a database was created. The total number of songs was about 600, of which about 150 were used in dances.

We thus try to extract the characteristics of Keihan's Zashiki dance worldview from the viewpoint of song words using text mining technique. The analysis can be summarized in the following two perspectives: difference in lyrics of dance songs and performance songs (characteristics), and differences in lyrics (characteristics) between different schools (iemoto).

We analyzed the frequency information of the vocabulary we used, focusing on nouns and adjectives, using RMeCab, an interface for using the Japanese morphological analyzer MeCab from the statistical analysis environment R. RMeCab is a text mining tool that can perform integrated morphological and statistical analysis processing. RMeCab is a text mining tool that integrates morphological analysis and statistical analysis processing.

Although no significant differences were observed in lyrics of dance songs and performance songs, for example, the adjectives “painful” and “happy” were used more frequently in the danced songs than in the other songs. Furthermore, hierarchical clustering of the collected text data into dance songs and performance songs using the Ward method confirmed that the data were roughly clustered into dance songs and performance songs, although there was some confusion in some parts.

Next, we attempted a cluster analysis of the characteristics (differences) of the lyrics of the pieces performed by the different schools (iemoto). The data of the performance pieces are the pieces listed in the “Cultural Digital Library” of the Japan Arts Council (the data of the voluntary performances of dance and traditional Japanese music at the National Theatre of Japan and the National Bunraku Theatre of Japan). The three clusters were neatly classified into three Yoshimura Schools, two Inoue Schools, and a Yamamura School, and the degree of similarity within the schools is clear. This allowed for the possibility of automatic classification using text mining techniques.

We confirm that each art has its own unique style such as: Inoue, who has been handed down only by women; Yoshimura style, which has been handing down women's dances in recent years while keeping men at home; and Yamamura style, which has a flow of Kabuki.

Keywords: Learning performing arts, dance song analysis, text mining, clustering analysis.