利用人工智慧技術偵測中文假新聞

Simple item page
dc.contributor曾元顯zh_TW
dc.contributor.author林郁綺zh_TW
dc.contributor.authorLin, Yu-Chien_US
dc.date.accessioned2022-06-08T03:01:18Z
dc.date.available2021-01-17
dc.date.available2022-06-08T03:01:18Z
dc.date.issued2021
dc.description.abstract在資訊快速傳播的時代，假新聞滿天飛的困境肆虐全世界，在資訊爆炸的時代如何使用資訊科技的技術快速過濾虛假的資訊是此研究想要探討的問題。 本研究為探討人類與電腦在中文假新聞偵測上的實際表現，分別以人類與電腦為出發點進行三個實驗，「自然語言模型辨別假新聞」根據臺灣假新聞平台「CoFacts 真的假的」建置中文假新聞語料，並包含知識推論標記，再使用Naïve Bayes、SVM以及BRET進行真假新聞預測；「人類辨別假新聞編寫模型」根據經濟日報語料，使用GPT2-Chinese生成假新聞，並請受試者辨別真假新聞；最後以「假新聞自動編寫模型評估」整合前兩項實驗，以分類器實測電腦是否能辨別出GPT2-Chinese自動生成的假新聞，並比較與受試者的差異，實驗結論如下： 1. BERT預測真假新聞MicroF1為0.8184，MacroF1為0.7686，顯示電腦在一定程度上能夠輔助人工辨別假新聞，但並非真正瞭解語意。 2. 受試者辨別GPT2-Chinese自動生成之假新聞，其真新聞平均可信度為3.68，假新聞為2.54，顯示閱讀者可以辨別真假，但不具有背景知識的受試者較難辨別，而新聞與受試者越相關越會提高轉發意願。 3. BERT預測問卷的30篇新聞，其MicroF1與MacroF1皆為0.93，僅2篇錯誤，而人類判斷錯誤為5篇，且判斷錯誤的新聞完全不重疊，顯示電腦可以辨別電腦所產生的假新聞，並且與人類有互補合作之處。 綜合而言，本研究的貢獻不僅建置了包含知識推論之假新聞語料庫，並進行分類器評測；且從反向思維實作了假新聞編寫模型之訓練，更以人類與電腦進行實測，奠定了未來假新聞研究之基石，期待日後能有更多研究者投入於此。zh_TW
dc.description.abstractIn the era of information explosion, fake news is raging over the world, how to use artificial intelligence technologies to distinguish fake news is this study exploring. To evaluate humans and computers in the detection of fake news in Chinese, this study conducts two experiments from different opinions. In the first experiment"Evaluation of Detection of Fake News by Natural Language Processing", classify Taiwan fake news website "CoFacts" by BERT; in the second experiment "Human Distinguish Fake News Write Model", generate fake news by GPT2-Chinese and interview reader to distinguish, the conclusions are as follows: 1. The BERT highest prediction rate is 79%. 2. Computers can help identifying fake news, but not by semantics. 3. Readers identify fake news by GPT2-Chinese, the average credibility of true news is 3.68,and the average credibility of fake news is 2.54. 4. Readers can distinguish between true and fake news, readers without economic background are more difficult to distinguish. This research not only builds and evaluates a fake news corpus with knowledge inference; also trains fake news writing models and evaluates, foundation for future fake news research. I'm looking forward to more researchers invest this in future.en_US
dc.description.sponsorship圖書資訊學研究所zh_TW
dc.identifier60715006E-38332
dc.identifier.urihttps://etds.lib.ntnu.edu.tw/thesis/detail/20dc10ec20440b5a66c1bac6258d26d9/
dc.identifier.urihttp://rportal.lib.ntnu.edu.tw/handle/20.500.12235/118316
dc.language中文
dc.subject假新聞偵測zh_TW
dc.subject人工智慧zh_TW
dc.subject假新聞語料zh_TW
dc.subject知識推論zh_TW
dc.subject文字生成zh_TW
dc.subjectFake News Detectionen_US
dc.subjectArtificial Intelligenceen_US
dc.subjectFake News Corpusen_US
dc.subjectKnowledge Inferenceen_US
dc.subjectWriting Modelen_US
dc.title利用人工智慧技術偵測中文假新聞zh_TW
dc.titleExploring Artificial Intelligence Technologies for Fake News Detectionen_US
dc.type學術論文

Files

Original bundle
Now showing 1 - 1 of 1
No Thumbnail Available
Name:
60715006E-38332.pdf
Size:
2.33 MB
Format:
Adobe Portable Document Format
Description:
學術論文
Download

Collections

學位論文