本研究針對風災新聞報導，區分早期與近期新聞兩個時期，嘗試從報導框架語言建構、框架核心意義與報導趨勢等三方面檢視這兩個時期處理風災災前報導的差異。本研究採取質量化混合方法，利用文字探勘方法與視覺化來抽取詞彙關係、比較用字差異，以及抽取相關文本來外化框架分析的過程。透過自動化文字探勘輔助文本分析比較後發現，這兩個時期在新聞框架的呈現上由早期的「應變自然」過渡到近期的「社會應變」。前者著重自然災情的資訊傳達，後者則把關注的焦點擴及到社會整體，包含了專家角色、社會制度，與個別／集體行動者的行動力等。本研究提供具有可驗證性的實證資料，並使用可重置性的文字探勘方法輔助偵測或搜尋目標文本，期望作為未來應用資料科學方法於質量化混合研究的參考。

This study investigated early and recent news reports on typhoons and typhoon-related disasters released by United Daily News from the perspectives of language construction, core meanings, and reporting trends. This study collected news reports from January 1, 1986 to December 31, 1996, and from January 1, 2006 to December 31, 2016, representing early and recent news respectively. The study employed text mining and frame analysis methods and studied the “frame” difference of news media over the past three decades. An analysis of the language construction of the news reports revealed that, concerning the subject of social communication, recent typhoon reports often utilized opinions from different social “actors” to increase the conflicts and violations of norms and human rights. From early to recent period, it is found that news frame are transiting from the early “response to nature” to the recent“response to society”. The former focuses on the information transmission of natural disasters, while the latter expends the focus of attention to the whole society, including the role of experts, social systems, and the actions of individual or collective actors. We believed that the study provides verifiable empirical data, uses text mining methods to assist in detecting or searching target texts, automates the detection of topics, and finds out important keyword pairs, and hopes to serve as a reference for future applied data science methods in qualitative hybrid research.

文字探勘；主題建模；詞嵌入；框架分析；災難新聞研究

Text mining; topic modeling; word embedding; frame analysis; disaster news researches