誠實標題黨：具吸引力且忠於事實的新聞標題產生器

Abstract

隨著網際網路與社群媒體的興起，資訊產生及傳播的速度也不停的在增長。假新聞儼然成為了這個世代重要的議題之一，而其中一種對抗假新聞的方法，就是撰寫澄清新聞來核實不正確的資訊。然而澄清新聞的主要目的在於闢謠，使用的口吻時常過於平淡，容易導致讀者喪失興趣，而使點閱率與假新聞相比較低。 與此同時，深度學習的發展不斷的在縮小諸多任務中機器與人之間的距離，語言模型的成熟使得自動生成文章的摘要或標題變得可能，許多研究也以此為方向，希望能夠讓機器來幫助人們撰寫文案。過去的研究主要以點擊率為依據，也是判斷一個新聞是否具有吸引力的唯一指標，卻可能忽略了新聞事件本身也可能是造成點擊率高的原因之一，因此若以這樣的標準收集資料並訓練模型，反而可能使得真正具有吸引力的標題成為噪音而影響模型的表現。 在這個研究中，我們先透過讀者研究，分析具吸引力的標題所具備的風格，及其在真假新聞之間的差異；接著我們讓模型透過假新聞資料學習出產生具吸引力標題的能力，再計算產生標題的聳動程度及真實程度，並以強化式學習的方法來更新整個框架。實驗結果顯示我們的方法能夠在吸引力、真實性取得顯著的進步，並在不損失流暢性的情況下擊敗多個過去最優的語言生成模型。

The dissemination of fake news has already become a major issue in this century, thanks to the rapid growth of the internet and social media platforms. One typical strategy for combating fake news is to release verified news. However, most verified news uses a monotonic tone to point out the fact, which loses readers interest and thus being less effective. Current methods for generating attractive headlines often learn directly from data, which bases attractiveness on the number of user clicks and views. Although clicks or views do reflect user interest, they can fail to reveal how much interest is raised by the writing style and how much is caused by the event or topic itself. Also, such approaches can lead to harmful hallucinations by over-exaggerating the content, aggravating the spread of false information. In this work, we propose HonestBait, a novel framework for solving these issues from another aspect: generating headlines using forward references(FRs), a writing technique often used in clickbait. A self-verification process is also included to avoid harmful hallucinations. Automatic metrics and human evaluations show our framework yields better results in attractiveness while maintaining high veracity.