근대 잡지 『삼천리』의 의미 군집을 찾아서-비지도 학습을 활용하여-

초록

This paper aims to get representative semantic clusters of whole articles in 『Samcholli』 (the whole land of Korea) which is a magazine published during the modernization period in the Japanese colonial era. Rather than working deductively to group words from articles into certain categories, we work inductively to group words from articles into latent topic word groups as we call them semantic clusters. First, we collect all the articles of 『Samcholli』 made available by the National Institute of Korean History. Second, taking word & token analysis to assess semantically meaningful words, we experiment on all semantic token sets by three clustering methods, KMeans, LDA topic modeling, and Word2vec embedding as unsupervised learning. Thirdly, we show that 『Samcholli』 covered a great deal of societal aspects related to the modern economic system, tuberculosis, the second world war, modern art, and films which have not well yet been dealt with in literature studies. Finally, we argue that these word representative learning methods of practically achieving the aspects of the societal kaleidoscope of 『Samcholli』 are a great tool for enriching and extending qualitative analysis and views of prior works of literature.

키워드