Stephen Colbert at SemEval-2023 Task 5: Using Markup for Classifying Clickbait

Proceedings of the The 17th International Workshop on Semantic Evaluation (SemEval-2023), pages 1844–1848 July 13-14, 2023 ©2023 Association for Computational Linguistics Stephen Colbert at SemEval-2023 Task 5: Using Markup for Classifying Clickbait Sabrina Spreitzer University of Regensburg Sabrina.Spreitzer@student.ur.de Hoai Nam Tran University of Regensburg Hoai-Nam.Tran@student.ur.de Abstract For SemEval-2023 Task 5, we have sub- mitted three DeBERTaV3LARGE models to tackle the first subtask, classifying spoiler types (passage, phrase, multi) of clickbait web articles. The choice of basic parame- ters like sequence length with BERTBASE un- cased and further approaches were then tested with DeBERTaV3BASE only moving the most promising ones to DeBERTaV3LARGE. Our re- search showed that information-placement on webpages is often optimized regarding e.g. ad- placement. Those informations are usually de- scribed within the webpages markup which is why we conducted an approach that takes this into account. Overall we could not manage to beat the baseline, which we lead down to three reasons: First we only crawled markup for Huffington Post articles, extracting only <p>- and <a>-tags which will not cover enough aspects of a webpages design. Second Huff- ington Post articles are overrepresented in the given dataset, which, third, shows an imbalance towards the spoiler tags. We highly suggest re-annotating the given dataset to use markup- optimized models like MarkupLM or TIE and to clear it from embedded articles like "Yahoo" or archives like "archive.is" or "web.archive" to avoid noise. Also, the imbalance should be tackled by adding articles from sources other than Huffington Post, considering that also multi-tagged entries should be balanced towards passage- and phrase-tagged ones. 1 Introduction The shared task "Clickbait Challenge at SemEval 2023 - Clickbait Spoiling" on PAN was about classi- fying and predicting spoilers from English (gossip) articles (Fröbe et al., 2023a) provided via TIRA (Fröbe et al., 2023b). It poses several technical challenges, especially handling text data with a high degree of diversity in genre, writing style, and structure. The task aims to close the curiosity gap clickbait posts cause within their readers by provid- ing an informative summary. Solving this problem would benefit the use of social media since click baits are an annoying phenomenon aiming to ma- nipulate one into visiting a page which may affect the user’s credibility and quality perception in a negative way (Hagen et al., 2022). The task con- sists of two subtasks: 1. classifying spoilers into three categories: "pas- sage," "phrase," or "multi" 2. predicting spoilers for an article based on those categories Our system focuses on solving the classification task with a fine-tuned DeBERTaV3LARGE model trained on the given Webis Clickbait 2022 Corpus, which we enriched with information from markup crawled from the original websites. We have cho- sen this approach since we discovered that (gossip) articles, in most cases, as webpages, are subject to some professional style guide aiming to make visi- tors stay on the page, which is essential, especially when it comes to advertising revenues (Nielsen and Pernice, 2009). The results fell short of ex- pectations since we only tested …

Content truncated to 500 words