Alexander Knox at SemEval-2023 Task 5: The comparison of prompting and standard fine-tuning techniques for selecting the type of spoiler needed to neutralize a clickbait

Proceedings of the The 17th International Workshop on Semantic Evaluation (SemEval-2023), pages 1470–1475 July 13-14, 2023 ©2023 Association for Computational Linguistics Alexander Knox at SemEval-2023 Task 5: The comparison of prompting and standard fine-tuning techniques for selecting the type of spoiler needed to neutralize a clickbait Mateusz Wo´zny and Mateusz Lango Institute of Computer Science Poznan University of Technology Poznan, Poland mateusz.wozny@student.put.poznan.pl Abstract Clickbait posts are a common problem on so- cial media platforms, as they often deceive users by providing misleading or sensational headlines that do not match the content of the linked web page. The aim of this study is to cre- ate a technique for identifying the specific type of suitable spoiler - be it a phrase, a passage, or a multipart spoiler - needed to neutralize click- bait posts. This is achieved by developing a machine learning classifier analyzing both the clickbait post and the linked web page. Mod- ern approaches for constructing a text classifier usually rely on fine-tuning a transformer-based model pre-trained on large unsupervised cor- pora. However, recent advances in the devel- opment of large-scale language models have led to the emergence of a new transfer learn- ing paradigm based on prompt engineering. In this work, we study these two transfer learning techniques and compare their effectiveness for clickbait spoiler-type detection task. Our ex- perimental results show that for this task, using the standard fine-tuning method gives better re- sults than using prompting. The best model can achieve a similar performance to that presented by Hagen et al. (2022). 1 Introduction 1.1 Problem description Clickbaits are social media posts designed to grab the reader’s attention in order to visit a linked web page. As they are used by publishers to increase the number of views, clickbait posts are becoming more and more common in the webspace. Even though some clickbait posts are harmless, a significant amount of them is misleading and help spread fake news. Therefore, some research works have been devoted to clickbait detection problem (Potthast et al., 2016), leading e.g. to the proposal of an additional plug-in for the browser to alert the user of a clickbait header (Chakraborty et al., 2016). However, clickbait detection alone is not enough, as it only identifies the problem, but does not solve it. Therefore, Hagen et al. (2022) proposed using automatically generated spoilers to overcome the issue of clickbait posts. Spoilers are short texts that provide the most necessary information from the web page linked to the clickbait, completely neu- tralizing the clickbait post. Three types of spoilers are put forward (Hagen et al., 2022): • phrase – a single word or phrase from the linked page • passage – one or a few sentences from the linked page • multipart – a few non-consecutive phrases or passages extracted from the linked page For instance, for a given clickbait post ”Obama Admin Just Outlawed Wildly Popular Product Found in Millions of Bathrooms”, the correspond- ing spoiler is a phrase ”antibacterial soaps” con- structed from …

Content truncated to 500 words