機器學「習」：以文字探勘法探索習近平時期之大外宣戰略

Machine Learning: An Application of Text Mining to Xi's Grand External Propaganda Strategy

邵軒磊(Hsuan-Lei Shao)

《中國大陸研究》 62卷4期 (2019/12) Pp. 133-157

https://doi.org/10.30389/MCS.201912_62(4).0005

摘要

因為中共政經地位的轉變及其「銳實力」的影響，國際上之「中國形象」在近年有相當變化。尤以習近平提出「大外宣」戰略後，上述發展更為顯著，同時其細節也尚待研究。對於研究者而言，因為各種資料的質與量的快速成長，使得以個人經驗與智識判斷為主的研究方式受到挑戰。相對於此，使用數位方法的優點在於得以檢驗並更有效累積；觀察資料與模型，也能適應未來變化而增添調整。因此，本文作者試圖將數位技術應用至此一題目，比如文字探勘、機器學習、主題分析模型等，在龐大現代政治論述文本中建立主題模型，嘗試尋找政治領袖在演講中所透露之政治訊息以及政治價值，亦即指出習近平時期演講在各個主題的概括樣貌。初步的解答是：以習近平自身講話做主題分析後，確實發現其對外與對內用語不同，機器能分辨並歸類其用語特色；也能看出在領域上則「外交、經濟、生態」類文本之主題與「黨建、政治、國防」之主題不同。本研究蒐集了中國國家主席習近平的發言作為語料庫，並使用數位方法初探中文政治文獻，期待藉此關注中共大外宣與銳實力。

關鍵字

文字探勘 ； 機器學習 ； 大外宣 ； 習近平

並列摘要

＂China's image＂ has undergone a dramatic transformation as China keeps expanding its international influence through its rising sharp power. In particular, much of the propaganda work under Xi Jinping has been carried out through a ＂Grand External Propaganda Strategy＂. Growing concerns about the implications of this strategy demand a closer look at its characteristics and practice. Yet, there is a major methodological challenge of analyzing massive amounts of data efficiently and accurately while not relying solely on personal understandings. To solve this challenge, digital methods are required, which will be the topic of this paper. This research collated Xi's speeches into a database and developed digital tools to process and analyze ＂China's Sharp Power＂ and ＂Grand External Propaganda Strategy.＂ It uses text mining and machine learning, as well as adopting the ＂Latent Dirichlet Allocation＂ (LDA) model to extract the main themes from various propaganda texts in order to identify the political messages and ideologies the Chinese top leadership tried to communicate. This paper demonstrates that there are significant differences between speeches on domestic topics: DIPLOMACY, ECONOMICS and ECOLOGY, and on externally directed topics: PARTY-BUILDING, POLITICS and DEFENCE. This research collected Xi's massive political speeches to a database, developed digital tool to process them, and analyzed ＂China Sharp Power＂ and ＂Grand External Propaganda Strategy＂ to explore the unknown future.

並列關鍵字

Text Mining ； Grand External Propaganda Strategy ； Xi Jinping ； Machine Learning

參考文獻

人民網，〈習近平系列重要講話資料庫〉，http://jhsjk.people.cn/，查閱時間：2018/11/15。People.cn. 2018. “Xijinping xilie zhongyao jianghua ziliaku” [Xi Jinping’s series of important speech databases]. (Accessed on November 15, 2018).

何清漣，2019，《紅色滲透：中國媒體全球擴張的真相》，臺北：八旗文化。He Qing-liang. 2019. Hongse shentou: zhongguo meiti quanqiu kuozhang de zhenxiang [Red Infiltration: The Truth of Global Expansion of Chinese Media]. Taipei: Bachi Culture.

李彥冰、荊學民，2010，〈全球化背景下我國對外傳播內外有別原則的困境與出路〉，《湖南大眾傳媒職業技術學院學報》，10 (3)：29-33。Li, Yan-bing, Xue-min Jing. 2010. “Quanqiuhua beijing xia woguo duiwai chuanbo neiwai youbie yuanze de kunjing yu chulu” [Dilemma and Way-out of the Principle of Making Difference between Internal and External in International Communication in Globalization]. Journal of Hunan Mass Media Vocational Technical College, 10 (3): 29-33.

邵軒磊，2012，〈中國研究議程之系譜：以日本國際政治學會誌為例〉，《問題與研究》，51 (1)：23-54。Shao, Hsuan-lei. 2012. “Zhongguo yanjiu yicheng zhi xipu: yi riben guoji zhengzhixue huizhi weili” [Genealogy of the China Studies Agenda-In the Case of Journal Publications of the Japan Association of International Relations (JAIR)]. Issues and Studies, 51 (1): 23-54.

邵軒磊、曾元顯，2018，〈文字探勘技術輔助主題分析：以「中國大陸研究」期刊為例〉，《問題與研究》，57 (1)：29-62。Shao, Hsuan-lei, and Yuen-hsien Tseng. 2018.“Wenzi tankan jishu fuzhu zhuti fenxi: yi zhongguo dalu yanjiu qikan weili” [Text Mining-based Topic Analysis: A Case Study of the Journal "Mainland China Studies"]. Issues and Studies, 57 (1): 29-62.