從媒體微博之限制評論設置看中國網路輿論限縮

新浪微博等社群媒體的出現，徹底改變了中國傳統媒體壟斷已久的信息傳播路徑，為中國單一的輿論空間帶來豐富多元的聲音。中國媒體同樣感知到新浪微博等社群媒體帶來的影響力與機會，紛紛開設微博以擴大新聞傳散。但同時，中國政府亦在不斷適應互聯網的發展，改變互聯網管理策略以維護政治穩定。2018年初起，新浪微博平台提供給微博主「限制評論機制」，即可依據其目的選擇性地將部分貼文設置為限制評論。那麼限制評論的貼文具有怎樣的特色？不同媒體對貼文限制評論的操作又有何差異？其動機可能為何？而在新浪平台提供這樣的機制的環境下，媒體應用該設置限評貼文有可能會對公眾輿論的發展帶來什麼樣的影響？本研究立意爬梳《新華視點》、《人民日報》、澎湃新聞、《新京報》四家媒體微博賬號發佈的貼文與評論，並透過文字探勘方法對限評與非限評帖文內容進行分析。研究結果發現，對政治相關的貼文，媒體的自我審查更加警惕：含有報導國家領導與政黨的專用名詞、報導國家領導與政黨事件的情境詞彙、報導政府施行社會民生策略的關鍵詞之貼文，有更大比例被設置為限制評論。其次，透過限評主題抽取發現，與政治議題相關的貼文，有更大比例被設置為限制評論。不同媒體對貼文的限評設置存在區別，@新華視點的微博貼文幾乎是全面限評的策略，而@新京報則是將與政治議題相關的貼文設置為限制評論，其他議題之貼文則不設限制，給予閱聽眾最大的討論空間。透過反復爬取貼文評論來偵測貼文評論狀態的變更，本研究發現，刪除評論的操作普遍存在於本研究的研究對象中，而提及本研究所推估的高敏感詞彙之評論則幾乎被盡數刪除。從過往研究來看，新浪平台因應國家網路輿論政策方向，不僅從平台直接透過關鍵字篩選輿論，本研究中，其自2018年開始提供貼文限評機制，而媒體在平台所提供的限評選項下，因應媒體特性選擇自我審查之力度，對自家微博貼文做出限評設置，整體上恐限縮了閱聽眾與媒體、閱聽眾之間的輿論討論。

關鍵字

新浪微博；自我審查；限制評論；主題模型；文字探勘

並列摘要

As the emergence of social media may have great impact on the one-sided communication pattern in China, many traditional news agencies show great passion on posting news and talking with audience on Sina Weibo. However, for the purpose of censorship, since the beginning of 2018, editors of media weibo accounts have the privilege to choose which comments can be shown in public, thus the right to express freely on news turning to be shut down. This paper tend to find what kind of posts’s comment-status are more likely to be set as ‘limited’. The full content of posts, comments and comment-status of @新華視點，@澎湃新聞，@人民日報 and @新京報 within one year (from Feb. 2018 to Feb. 2019) were collected by crawler. Statistical method Chi-squared feature selection is used for selecting words that influence the comment-status of posts. Then the approach RR function () become highly effective in filtering words that only have impact on ‘limited’ comment-status posts. Result shows that most of the words are related to politicians or political affairs, which could also imply that posts with these selected words are more likely to be set into ‘limited’ comment-status. And also, by utilizing Structure Topic Model, corpus from those 4 media in Sina Weibo are classified as about 30-40 topics. Comparing the average gamma value of the two groups (limited and unlimited) in each topic, researcher find in some topics the average gamma value of ‘limited’ group are better than that in ‘unlimited’ group. Furthermore, the result of T-test on that difference between the two groups provide statistical proof. It is shown that posts in political or politics-related topics are more likely to be set as ‘limited’ comment-status. Editors not only choose which comments to be shown in public, but also delete some comments directly. After collecting all the comments for many times through the timeline during a short period, researcher find that the comment-status of posts were changed and some comments of posts were also omitted. This research reveals that the communication between media and audience has been restricted since Sina Weibo equipped editors with that privilege in 2018. To be more specific, in order to censor sensitive content and prevent them from publishing, Sina Weibo provides convenience for those media accounts and in return, those media weibo accounts use such convenience for self-censorship.