運用自然語言處理偵測台灣網路新聞媒體偏差

網路新聞是大眾重要的信息來源，至少已經部分替代了電視或印刷媒體等傳統媒體。但與其他媒體一樣，網路新聞媒體也可能受到媒體偏差的影響。媒體偏差是指記者或新聞編輯在選擇要報導的事件與報導的方式時造成的偏差，代表與新聞報導的標準產生了偏差。媒體偏差可能使媒體無法精準傳達事實並造成大眾對於媒體的不信任感，因此媒體偏差的認知與偵測是十分重要的議題。本研究使用既有研究中最具代表性的媒體偏差架構針對2020大選前三個月網路新聞媒體的政治新聞進行媒體偏差分析，為了分析這一龐大的資料集並比較不同網路新聞媒體的潛在政治傾向，我們提出了各種可以揭示媒體偏差的衡量指標。此外，本研究提出了一個可以從新聞文章中擷取出政治人物發言的方法與特徵級別情緒分析做法，兩者皆取得了不俗的準確度。最後根據分析結果，本研究發現台灣網路新聞媒體確實存在媒體偏差且大部分都擁有其潛在的政治傾向偏好。

關鍵字

媒體偏差；自然語言處理；網路政治新聞

並列摘要

Internet news is an important source of information for the public, and has at least partially replaced traditional media such as television or print media. However just like other media, online news media may also be affected by media bias. Media bias refers to the bias caused by the reporter or news editor in choosing the event to be reported and the reporting method, which represents a deviation from the standard of news reporting. Media bias may prevent the media from fairly and faithfully communicating facts and cause the public to distrust the media. Therefore, the recognition and detection of media bias is a very crucial issue. This study uses the most representative media bias framework in existing research to conduct media bias analysis on the political news of online news media three months before the 2020 presidential election in Taiwan. In order to analyze this huge data set and compare the potential political tendency of different online news media, we have put forward various metrics that can reveal media bias. In addition, this research proposes a method that can extract politicians' speeches from news articles and a feature-level sentiment analysis method, both of which have achieved satisfying accuracy. Finally, based on the analysis results, this study found that Taiwan’s online news media do exist media biases and most of them have potential political preferences.

並列關鍵字

Media Bias ； Natural Language Processing ； Internet Political News

參考文獻

Baly, R., et al. (2018). Predicting factuality of reporting and bias of news media sources. Proceedings of the 2018 Conference on Empirical Methods in Natural Language Processing.

Google Scholar

Baly, R., et al. (2019). Multi-Task Ordinal Regression for Jointly Predicting the Trustworthiness and the Leading Political Ideology of News Media. Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies.