運用大數據與GPT-3語言模型探究社會知覺下候選人之網路新聞報導與社群輿論分析—以2022年女性候選人為例

Abstract

在2022年九合一選舉中，台灣地方首長女性參選人數創下台灣30年來最高紀錄，於同年，各黨開始出現眾多正妹候選人參選現象，引起新聞媒體與社群媒體熱烈討論。因此，本研究將研究問題分為網路新聞、社群媒體兩個面向作探討。首先，探討網路新聞面向：在《聯合新聞網》、《自由時報電子報》 、《中時新聞網》 傳統三大報網路新聞媒體中，針對「2022女性里長候選人」相關之新聞主要關注面向為何？其次，探討社群媒體面向：在社群媒體PTT中，針對「2022女性候選人」相關之文章主要關注面向為何？將透過大數據分析取徑、詞頻分析與文字雲，對文本進行分析。最後，針對「2022女性候選人」相關文章下的留言來探討是否存在政治娛樂化現象？輔以GPT-3語言模型將PTT之留言分類為「政治娛樂化」、「非政治娛樂化」、「與選舉無關」三種類目。研究結果顯示，研究問題一結果：針對「2022女性里長候選人」之詞頻分析發現，三間網路新聞皆出現最美、正妹、美女詞彙，出現物化女性、將女性幼體化現象，也發現《自由時報電子報》較其他兩大報更娛樂化。研究問題二結果：針對「2022女性候選人」之PTT文章詞頻分析發現，「民進黨」與「國民黨」兩大黨為網民最關注的面向，而「柯文哲」一詞在PTT文章及留言為高頻率詞彙，發現PTT存有網軍現象。而從「2022女性候選人姓名」討論詞頻中也發現社群討論頻率與得票數不一定相關。研究問題三結果：PTT留言存在政治娛樂化現象，包含歧視女性候選人之外表、以其姓名作諧音嘲諷，及出現諷刺民進黨支持者等貶義詞，顯示網路的匿名性特質出現態度極化的攻擊性言論與社群媒體之厭女言論現象。In the 2022 nine-in-one election, the number of female candidates for local cacique in Taiwan set a record for Taiwan in 30 years. In the same year, various political parties have many pretty girls participating in the election, which caused heated discussions in the news media and social platforms. Therefore, this study divides the research questions into two aspects: online news and social media. First, explore the aspects of online news. In"udn.com", "ltn.com", "chinatimes.com" What are the main concerns of the three traditional online news media of the three major newspapers in their reports on "2022 female village chief candidates"? Secondly, discuss the orientation of social media: In PTT, what are the main focuses of articles related to "2022 female candidates"? Explore the text through big data analysis, word frequency analysis and word cloud. Finally, let’s explore whether the comments on articles related to “2022 Female Candidates” are politically entertaining? The research also use GPT-3 to classify PTT comment into three categories:"political entertainment", "non-political entertainment", and "not related to elections".

Results of Research Question 1：The word frequency analysis of "2022 female village chief candidate" found that the words "most beautiful", "hot girls" and "beautiful girls" appeared in the three online news, which appeared to objectify women and infantilize women.Among the three major newspapers, it was also found that "ltn.com" contained more entertainment-oriented words than the other two major newspapers.

Results of Research Question 2：Word frequency analysis of PTT articles on "2022 Female Candidates" found that the two major parties, DPP and KMT, were discussed most frequently by netizens, and "Ko Wen-je" is a high-frequency word in PTT articles and comments, and it is found that have cyber warrior on the PTT forum. "2022 Female Candidate Names" also is a high-frequency word, that found social opinion is not necessarily related to votes.

Results of Research Question 3: There are political entertainment phenomena in PTT comments, including discriminating against the appearance of female candidates, mocking their names with homophonic names, using derogatory terms to satirize DPP supporters, etc., which highlights the anonymity of the Internet and the popularity of social media platforms. There is widespread offensive rhetoric and misogynistic rhetoric.