Overview of PoliticHeadlinES at IberLEF 2026: Multimodal Headline Ranking in Spanish Political News

Resumen

This paper presents PoliticHeadlinES 2026, a shared task on headline ranking for Spanish political news. Given a news article and ten candidate headlines, systems must rank them by relevance, placing the original headline at the top. The competition comprised two subtasks: a text-only setting and a multimodal setting incorporating the article's image. The dataset contains over 20,000 political news articles from multiple Spanish media outlets, using real-world distractor headlines to create plausible alternatives. A total of 19 teams participated, submitting 168 runs. Results show that headline ranking remains highly challenging; the inclusion of visual information yielded limited improvements over text-only approaches. Systems leveraging large language models and reasoning-oriented strategies achieved the top overall performance, emphasizing the need for deep contextual and semantic matching.