機器學習與日本研究的對話：以中國社會科學院《日本學刊》為例

Dialogue of Machine Learning and Japan Studies: Analysis of Journal Japanese Studies of Institute of Japanese Studies, Chinese Academy of Social Sciences (JSCASS)

以往對某一人文知識領域的研究，往往受限研究者的個人時間與體悟靈感，如何有效率的使用時間，一直是人文學者的難題。但若加入跨領域思維，在現代科技的幫助之下，吾人可以使用數位工具對某一領域做出宏觀視角，幫助學者歸納「大量知識」。具體而言，本文嘗試結合跨領域研究手法，機器學習（Machine Learning, ML）與自然語言處理（Natural Language Processing, NLP）等，來處理《日本學刊》（中國社科院日本研究所學誌）之三十七年間二三三七篇論文，並歸納《日本學刊》研究論文的「發表類型」、各類代表性關鍵字有哪些，與各年的「發表取向」有什麼變化。本文成功實做大量中文長文本的文字解析，並能抽取其中資料訊息，並將流程半自動化，可以達到輔助研究者快速大量理解訊息。本文能抽取出高產作者、高產機構、歷年發表情形以及日本研究的主要七類範疇「金融經濟、政治、管理、文學思想、文化社會、國際關係、社會經濟」。本研究期望能提供讀者對於某個知識領域的遠程宏觀視角，以此跨領域方式研究「東亞文明」，將讓研究者開啟更多可能性。

Since studies on knowledge areas/ knowledge communities are often limited by our labor time, research cost/performance shall hence be prioritized/selected. However, our vision could be enhanced by new interdisciplinary thinking. In this article, we employed the methods of Machine Learning and Natural Language Processing to analyze 2,237 research articles. Those articles were taken from The Journal Japanese Studies published by Institute of Japanese Studies, Chinese Academy of Social Sciences (JSCASS) between 1991 and 2017. Results from the data analysis provided the answers to two questions, namely (1) what fields can be clustered in Japanese Studies, accompanied with their keywords, and (2) the number flows of the fields by year. This study analyzed thousands of Chinese long texts, and captured the key information semi-automatically, in order to efficiently support readers to approach complex information in a simplified manner. This study disclosed information on highly productive writers, highly productive institutes, their annual publications, and the main topics of Japanese Studies, which are Economics, Politics, Management Science, Literature/Ideology, Culture/Society, IR, and Social Economy. It could also provide a macro perspective in the ''East Asian Studies'' by new interdisciplinary thinking.