Francis Wilde at SemEval-2023 Task 5: Clickbait Spoiler Type Identification with Transformers

Proceedings of the The 17th International Workshop on Semantic Evaluation (SemEval-2023), pages 1890–1893 July 13-14, 2023 ©2023 Association for Computational Linguistics Francis Wilde at SemEval-2023 Task 5: Clickbait Spoiler Type Identification with Transformers Vijayasaradhi Indurthi and Vasudeva Varma Information Retrieval and Extraction Lab International Institute of Information Technology, Hyderabad vijayasaradhi.i@research.iiit.ac.in and vv@iiit.ac.in Abstract Clickbait is the text or a thumbnail image that entices the user to click the accompanying link. Clickbait posts hide the critical elements of the article and reveal partial information in the title, which arouses sufficient curiosity and mo- tivates the user to click the link. In this work, we identify the kind of spoiler given a clickbait title. We formulate this as a text classification problem. We finetune pretrained transformer models on the title of the post and build mod- els for the clickbait-spoiler classification. We achieve a balanced accuracy of 0.70 which is close to the baseline. 1 Introduction Clickbait-spoiling (Hagen et al., 2022) aims to counteract the manipulative tactics used by click- bait creators and to help people avoid wasting their time on low-quality content. This is achieved by revealing the "suspense" that has been deliberately created in the title by the authors of the post. With- out clickbait spoiling, the reader has to spend con- siderable time to satisfy his curiosity by finding the content which was promised in the title. By clickbait-spoiling, the user can easily skim through the key content without wasting valuable time. The Clickbait Spoiling task at SemEval 2023 (Fröbe et al., 2023a) consists of two subtasks, Spoiler Type Classification and Spoiler Generation. • Spoiler Type Classification: In this task, the goal is to classify a given clickbait post into one of the three kinds of spoiler types i.e phrase, passage or multi depending on the kind of text required to spoil the clickbait. • Spoiler Generation: In this task, the goal is to generate the actual spoiler for the clickbait post. Both tasks are presented in the English language. We attempted the first task which is Spoiler Type Classification. We use pre-trained transformer mod- els and finetune them on the postText to train a spoiler-type prediction model. We also experi- ment with pretrained transformer representations and train classical ML models on these represen- tations for classifying the spoiler type. Our final model, a vanilla transformer model finetuned on this dataset, achieves a balanced accuracy metric of 0.7 which is close to the baseline. 2 Background In the task for Clickbait Spoiler Type classification, the input is a tweet along with its associated meta- data. Table 1 lists the various fields available from the dataset. Table 2 lists few examples of clickbaits and their spoiler types. We formulate the problem of spoiler type identi- fication as a text classification task. We hypothesize that a human can easily predict the spoiler type of a clickbait by looking at the post of the clickbait i.e in our case, it is the postText associated with the tweet. We adopt the method followed …

Content truncated to 500 words