假健康新聞是情感病毒所造成的一種群體歇斯底里

Abstract

背景：錯誤訊息新聞對公共衛生構成重大威脅。目前的應變措施主要集中在事實查核，或近期的人工智慧相關的方法。然而，這些方法需依賴專家的協助。現有查核組織多以人工方式進行，查核速度遠不及錯誤資訊被快速傳散的速度。本研究的目標是透過假資訊和可信賴資訊的內容特徵區分，使讀者能夠自主識別錯誤資訊。

研究方法：這是一項 1:2 案例對照觀察研究，採用質性與量性混合性研究。案例選自2020年至2023年間台灣事實查核（TFC）核實錯誤的文章，對比文章選則可受信賴的資訊來源CH。運用中文語料庫進行質性分析，從樣本中找出常見的關鍵字詞。然後拿TFC和CHM兩群文章各前50個高頻率的關鍵詞，接著在每篇文章計算這100個關鍵詞的頻率和密度。最後使用t檢定和單一邏輯迴歸對關鍵字密度進行兩群文章的量化比較。

研究結果：總共從TFC和CHM中分別選出了37篇和75篇文章。TFC文章明顯比 CHM文章短（225.08 +/- 201.75，vs 1409.04 ± 557.30，p<0.001）。排除與 COVID 相關的關鍵詞後，有7個關鍵詞在TFC文章中出現的密度明顯高於CHM文章。關鍵詞上下文的脈絡分析發現，這些關鍵詞通常具有情感作用。

結論：假新聞中包含的情感或不特定對象統稱的關鍵詞密度較高。假新聞可能是「情感病毒」，它的目的可能不是傳播知識內容，而是傳播情感。假新聞也可能是一種「知識病毒」，透過情感的「尖刺狀蛋白質」將錯誤資訊附著在讀者心智上。錯誤資訊的情感內容可能造成「群體歇斯底里」，形成一個新的公共衛生問題。應對假新聞的應變措施可能需要從群眾心理問題的角度，來解決資訊內容管理的問題。

Introduction: Misinformation is a significant health threat. Current interventions mostly rely on fact-checking or, more recently, on artificial intelligence-related approaches. The current study aims to empower readers to detect fake news for themselves by identifying the content characteristics that differentiate fake health-related news from reliable sources.

The goal of the current study is to empower readers to detect fake news for themselves by identifying its characteristics.

Methods: A 1:2 case-control observational study from 2020-2023 with mixed methods analysis was performed in Taiwan, which has been reported to have a high prevalence of fake news due to geopolitical tension. Cases were selected from articles confirmed to be false by Taiwan Fact Check (TFC), and comparison articles were selected from Common Health Magazine (CHM), generally considered the most reputable source of health information in Taiwan. Qualitative analysis was performed to identify common keyphrases. Frequency and density of the top 50 keyphrases from TFC and CHM were then calculated for every article, and quantitative comparison of keyphrase densities were performed with t tests and single logistic regression. Qualitative analyses were then performed for important keyphrases to determine their context.

Results: A total of 37 and 75 articles were selected from TFC and CHM, respectively. TFC articles was significantly shorter than CHM articles (225.08+/-201.75, vs 1409.04+/- 557.30, p<0.001). After excluding COVID-related keyphrases, 7 keyphrases appeared with significantly higher density in TFC articles than in CHM articles. Context analyses revealed that these keyphrases are generally non-specific and emotive.

Discussion: Fake news includes a higher density of non-specific and emotive language. This information may empower readers to detect fake news for themselves. Societal interventions may also need to focus on the emotional component of fake news, through reassurance and stress management training.