Clickbait Spoiling via Question Answering and Passage Retrieval

Proceedings of the 60th Annual Meeting of the Association for Computational Linguistics Volume 1: Long Papers, pages 7025 - 7036 May 22-27, 2022 c⃝2022 Association for Computational Linguistics Clickbait Spoiling via Question Answering and Passage Retrieval Matthias Hagen1 Maik Fröbe1 Artur Jurk1 Martin Potthast2 1Martin-Luther-Universität Halle-Wittenberg 2Leipzig University Abstract We introduce and study the task of clickbait spoiling: generating a short text that satisfies the curiosity induced by a clickbait post. Click- bait links to a web page and advertises its con- tents by arousing curiosity instead of provid- ing an informative summary. Our contribu- tions are approaches to classify the type of spoiler needed (i.e., a phrase or a passage), and to generate appropriate spoilers. A large-scale evaluation and error analysis on a new corpus of 5,000 manually spoiled clickbait posts— the Webis Clickbait Spoiling Corpus 2022— shows that our spoiler type classifier achieves an accuracy of 80%, while the question an- swering model DeBERTa-large outperforms all others in generating spoilers for both types. 1 Introduction Clickbait is the term used to describe posts in social media that are intended to inappropriately entice their readers to visit a web page. This is achieved through formulations such as sensationalism or cat- aphors that are believed to create a so-called cu- riosity gap: “a form of cognitively induced depri- vation that arises from the perception of a gap in knowledge or understanding” (Loewenstein, 1994). Clickbait is perceived as inappropriate since its res- olution is usually ordinary or trivial, comprising little more than a phrase, short passage, or a list of things that could just as easily have been included in the post. This observation motivates us to intro- duce the task of clickbait spoiling: identifying or generating a spoiler for a clickbait post. Figure 1 shows four examples of clickbait on Twitter, along with spoilers. The first two tweets explicitly or implicitly promise a surprising resolu- tion to spark curiosity, but their spoilers are brief and trivial. The linked page of the first tweet adds almost nothing, and the spoiler of the second is common sense. The third spoiler is a passage from the linked page, and the fourth is a list of things. New York Post @nypost Just how safe are NYC's water fountains? nyp.st/2yHSGnr “The Post independently tested eight water fountains in New York City’s most frequented parks, and found that all met or exceeded the state’s guidelines for water quality.” Lifehacker @lifehacker How to keep your workout clothes from stinking: lifehac.kr/57YOuEZ “washing [them]” Above the Law @atlblog The Surprising Way Recent Law School Graduates Are Getting Their First Job bit.ly/2CMMPxf “Networking.” Spoiler Clickbait tweet CNBC @CNBC A Harvard nutritionist and brain expert says she avoids these 5 foods that "weaken memory and focus." (via @CNBCMakeIt) cnb.cx/2TG6zeX “1. Added sugar” [...] “2. Fried foods” [...] “3. High-glycemic-load carbohydrates” [...] “4. Alcohol” [...] “5. Nitrates” [...] Figure 1: Examples of clickbait tweets and spoilers for them extracted from the respective linked web page. Even though there are length limits to …

Content truncated to 500 words