A Quantitative analysis into Waka collection and an Approach by machine learning

Abstract

　In this paper, as a part of text mining, we attempted to approach “Waka”, which is representing medieval literature. The target of analysis is eight imperial-commissioned waka anthologies which is called “Hachidaishu” from the middle of Heian period to the early of the Kamakura period. We performed a quantitative analysis using morphemes and examined the difference of “Hachidaishu”. We investigated as a method of analysis, using statistical difference of part-of-speech usage rate among “Hachidaishu”, especially seasonal poems and love poems, using the relationship between noun rate and MVR performed by Yumi Fujiike. Then, we tried to discriminate by using the Naïve-Bayes method which is one of machine learning methods. The result was that it was difficult to discriminate using the data of Waka overall. However, the discrimination of “Waka” by genre-specific data such as seasonal poems and love poems of “Waka” has also quite high one especially in “Gosen-Shu”. Therefore, it seems that“Gosen-Shu” is a little heterogeneous existence among “Hachidaishu ”.