基於圖神經網路之假新聞偵測研究

Abstract

在現今互聯網時代，隨著網路技術不斷發展，使得我們在閱覽資訊上也越來越便利，但與此同時，假新聞也藉著技術發展的順風車變得更容易生產傳播以及造成影響。所以本人便打算做和假新聞辨識相關的研究，找到較好的假新聞辨識的方法並提升假新聞辨識的準確率。由於看到CompareNet這一圖神經網路模型對於假新聞的辨識相較其他基礎方法效果更好。因此本研究是以CompareNet這一研究為基礎，基於LUN（Labeled Unreliable News Dataset）語料庫中的LUN-train語料庫創建了一個含有普通名詞、複數名詞、專有名詞、動詞、形容詞、副詞的知識庫，並將該知識庫和CompareNet這一研究相結合，使用LUN語料庫中的LUN-train語料庫來訓練模型、使用SLN（Satirical and Legitimate News Database）以及LUN語料庫中的LUN-test語料庫來對模型進行測試，提升假新聞辨識的準確率。

In the current internet era, with the continuous development of internet technology, accessing information has become increasingly convenient. However, at the same time, fake news has taken advantage of this technological advancement, making it easier to produce and spread, causing significant impacts. Therefore, I intend to conduct research related to fake news detection to find better methods for identifying fake news and improve the accuracy of fake news detection.After observing that the CompareNet graph neural network model has shown better results in fake news detection compared to other baseline methods, my work is based on the CompareNet study. I created a knowledge base containing common nouns, plural nouns, proper nouns, verbs, adjectives, and adverbs based on the LUN-train corpus (Labeled Unreliable News Dataset). Then, my work integrated this knowledge base with the CompareNet research. Trained the model using the LUN-train corpus and tested it using the SLN (Satirical and Legitimate News Database) and the LUN-test corpus from the LUN dataset to enhance the accuracy of fake news detection.