Welcome to the shared task TA1C (Te Ahorré Un Click) Clickbait Detection and Spoiling in Spanish, a task to automatically detect instances of clickbait in social media and generate short texts that spoil the clickbait. This task is part of IberLEF 2025.

Introduction

Clickbait is a widespread phenomenon in online news: it is a way of creating headlines and teasers aimed at capturing readers’ attention in order to increase traffic, relegating the function of informing to a secondary role. There is no clear consensus at the moment about how to define clickbait exactly, with some contradictory definitions that usually are based on the deceptive effect created by the news failing to deliver what they promise, or content based related phenomena such as sensationalism or yellow journalism. For this task we will take the following definition, based on Loewenstein's information gap theory (Loewenstein, 1994): “Clickbait is a method for generating teasers, especially online, that deliberately omits part of the information with the goal of generating curiosity by creating an information gap, thereby attracting the readers' attention and making them click” (Mordecki et al., 2024).

This work proposes a clickbait detection task that reduces the subjectivity of the decision by strictly following this definition and the identification criteria described in (Mordecki et al., 2024). It is also, and as far as we know, the first shared task on clickbait detection in Spanish. The corpus was created with the objective of representing as many Spanish varieties as possible, featuring news from media in 12 different countries as well as international media.

Although clickbait started in low-reputation web-exclusive media that focused on political propaganda or soft-news, such as The Huffington Post, Buzzfeed and Upworthy (Wu, 2016), it has gained prominence across all types of news and media. Relevant, hard-news are presented as clickbait more and more often nowadays. As a result, early intentions of detecting clickbait and automatically hiding it (Chakraborty et al, 2016) would imply missing important information. However, clickbait is usually perceived as annoying and it can lead to misinformation.

Spoiling the clickbait involves satisfying the curiosity by answering the information gap created. This way, the reader has all of the information and can decide to read the complete article based on interest and not curiosity, just as if the headline was written in a traditional way.

There are different kinds of information to be searched on to fill a clickbait related information gap, as described by (Fröbe et al., 2023). Many times it is a factual, specific piece of data, like a name, number or binary answer, or even an image or a list of answers (particularly for the case of listicles, the list articles popularized by Buzzfeed). Sometimes a whole summary is required, and in other cases there is no answer in the article at all.

There are some recent efforts on the clickbait spoiling task, mostly in English, but also in other languages such as Indonesian (Maharani et al., 2023) and, most relevant to this task, Spanish (García-Ferrero et al., 2024). There was a clickbait spoiling task in Semeval 2023 (Fröbe et al., 2023), where they focused exclusively on extractive spoilers (“spoilers in the form of text from the linked document”) and proposed three types of responses: a phrase, a passage, or having multiple parts.

This task is addressed to the Natural Language Processing community, especially (but not restricted to) those researchers working on the Spanish language, as well as to social science researchers, in particular journalism and communication. We anticipate that collaborative work may be particularly useful for this task, since clickbait is a relatively new phenomenon that has been changing, both in its characteristics and in its acceptance in the journalistic community.

Task description

We propose two subtasks for this competition about clickbait in Spanish:

Task 1: Clickbait Detection:

Determine if the content of a tweet that links to a piece of news is clickbait, given the definition of clickbait described above. Both the tweet and the news article are in Spanish. It is a binary classification problem, and the metrics for this task are accuracy, precision, recall and F1-score, with the latter being the main one. As the task has a level of subjectivity, we may ask the participants to indicate a clickbait score in addition to the binary label, and using those scores we will compute the Average Precision.

Task 2: Clickbait Spoiling:

Given a clickbait teaser (tweet and title) and the corresponding news article, the task consists in generating or extracting from the article a short text that, as concisely as possible (280 characters max), fills the information gap, satisfying the generated curiosity, or otherwise indicates that the articles has no response for it. The generated text must be in Spanish.

As this is a generative task and there is certainly more than one possible correct answer, we will use the following evaluation protocol: During the competition standard metrics like BERTScore (Zhang et al., 2020), BLEU-4 (Papineni et al. 2002) and ROUGE-L (Lin, 2004) will be used to compare between participants. After the test phase is over, we will perform a human evaluation of a subset of responses from the top 10 systems (according to the automatic metrics) and re-rank them according to a manual evaluation.

The human evaluators will rate each text according to its fluency, accuracy/completeness and conciseness in a 1 to 5, and we will average across different raters. This last human evaluation will not be revealed directly to the participants, as it will be done after the competition phases are finished. Here is a definition of the proposed metrics for the human evaluation:

Data

For task 1, we will publish a corpus of 3500 tweets (Mordecki et al. 2024), split 2800 for train and 700 for dev, collected throughout a year between October 2020 and October 2021 from 18 well known media outlets in Spanish, including many geographic varieties. Each tweet was labeled by three independent annotators, an contains the URL and clean HTML of the linked news, alongside the scraped headline, subheadline, clean text (only the news body), images with their captions and embedded URLs (usually social media links). The test set will comprise a new previously unpublished set of 700 tweets.

For task 2, we will manually write clickbait spoilers for at least 300 randomly sampled tweets from the task 1 dataset, as at least 100 more from the test set. These sets will be as balanced as possible regarding the different types of information gap that could exist in the data (e.g. factoid answers, lists, summaries, or no answer).

Examples:

Example 1

Example 2

Example 3

Example 4

Important Dates

Results

Official results will be published after the competition ends.

Contact

We will use the Codalab platform to manage participants and submissions. If you have any further questions, you can contact us via email.

The organizers of the task are:

Bibliography

(Chakraborty et al, 2016) Chakraborty, A., Paranjape, B., Kakarla, S., & Ganguly, N. (2016, August). Stop clickbait: Detecting and preventing clickbaits in online news media. In 2016 IEEE/ACM international conference on advances in social networks analysis and mining (ASONAM) (pp. 9-16). IEEE.

(Fröbe et al., 2023) Fröbe, Maik, et al. "SemEval-2023 task 5: Clickbait spoiling." Proceedings of the 17th International Workshop on Semantic Evaluation (SemEval-2023). 2023.

(García-Ferrero & Altuna, 2024) Iker García-Ferrero, Begoña Altuna (2024). NoticIA: A Clickbait Article Summarization Dataset in Spanish

(Lin, 2004) Lin, Chin Yew. (2004). ROUGE: A Package for Automatic Evaluation of Summaries. Text Summarization Branches Out, 74–81. https://aclanthology.org/W04-1013

(Loewenstein, 1994) Loewenstein, G. (1994). The psychology of curiosity: A review and reinterpretation. Psychological Bulletin, 116(1), 75–98.

(Maharani et al., 2023) Maharani, N. P. I., Purwarianti, A., & Aji, A. F. (2023). Low-Resource Clickbait Spoiling for Indonesian via Question Answering. In 2023 10th International Conference on Advanced Informatics: Concept, Theory and Application (ICAICTA) (pp. 1-6). IEEE.

(Mordecki et al., 2024) Gabriel Mordecki, Guillermo Moncecchi, Javier Couto. (2024) Te Ahorré Un Click: A Revised Definition of Clickbait and Detection in Spanish News. In Proceedings of Iberamia 2024.

(Papineni et al. 2002) Papineni, K., Roukos, S., Ward, T., & Zhu, W.-J. (2002). Bleu: a Method for Automatic Evaluation of Machine Translation. In P. Isabelle, E. Charniak, & D. Lin (Eds.), Proceedings of the 40th Annual Meeting of the Association for Computational Linguistics (pp. 311–318). doi:10.3115/1073083.1073135

(Wu, 2016) Wu, T. (2016). The Rise of Clickbait. In The Attention Merchants (pp. 276–289). Knopf.

(Zhang et al., 2020) Zhang, T., Kishore, V., Wu, F., Weinberger, K. Q., & Artzi, Y. (2020). BERTScore: Evaluating Text Generation with BERT. International Conference on Learning Representations. https://openreview.net/forum?id=SkeHuCVFDr