Search

The influence of fake news in the perception of reality has become a mainstream topic in the last years due to the fast propagation of miss-leading information, which has been enhanced by social media. To contribute to the fight against misinformation, researchers have proposed to develop automated solutions. The task of automated claim verification consists in assessing the truthfulness of a claim by finding evidence about its veracity. Datasets with synthetic claims have been developed to train models that perform this task. However, naturally-occurring claims are usually semantically more complex than synthetic claims. In this work, we test if the use of explicit semantic structures can help with the task of claim verification. We integrate Semantic Role Labels and Open Information Extraction structures to a BERT model, showing some improvement on the performance of the task. Additionally, we perform some explainability tests which show that the semantically-enriched model is better at handling complex cases, such as sentences in passive form or with multiple propositions.

Controversy is a social phenomenon that emerges when a topic generates large disagreement among people. In the public sphere, controversy is very often related to news. Whereas previous approaches have addressed controversy detection, in this work, we propose to predict controversy based on the title and content of a news post. First, we collect and prepare a dataset from a Spanish news aggregator that labels the news’ controversy in a community-based manner. Next, we experiment with the capabilities of language models to learn these labels by fine-tuning models that take both title and content, and the title alone. To cope with data unbalance, we undergo different experiments by sampling the dataset. The best model obtains an 84.72 micro-F1, trained with an unbalanced dataset and given the title and content as input. The preliminary results show that this task can be learned by relying on linguistic and social features. La controversia es un fenómeno social que ocurre cuando un tema genera desacuerdo entre los ciudadanos. En la esfera pública, la controversia se encuentra a menudo relacionada con las noticias de actualidad. Mientras que trabajos anteriores investigaron la detección de la controversia, en este trabajo nos proponemos predecirla basándonos en el título y el contenido de una noticia. En primer lugar, recogemos y curamos un conjunto de datos de un agregador de noticias en castellano que etiqueta las noticias según su controversia mediante las interacciones de la comunidad. Entonces, experimentamos con las capacidades de los modelos de lenguaje para aprender la categoría de controversia mediante el fine-tuneado de modelos que tienen el título y el contenido como contenido de entrada, y también con solo el título. Para lidiar con el desbalanceo de los datos, realizamos experimentos de sampleado de los datos. El mejor modelo obtiene una micro-F1 de 84.72, entrenado con un conjunto de datos desbalanceado y con el título y el contenido como entrada. Los resultados preliminares muestran que esta tarea puede ser aprendida mediante características lingüísticas y sociales. This work has been funded by the Spanish State Secretariat for Digitalization and Artificial Intelligence (SEDIA) within the framework of the Plan-TL, and the IBERIFIER project funded by the European Union (action number 2020-EU-IA-0252).

Previous studies indicate that the capacity of media to influence the salience of issues in the public realm is strongly dependent on specific attributes that characterize these issues. In this work, we investigate two internal aspects of issue types related to the attribute of duration. First, we address whether news stories belonging to different issue types can be identified and represented using a set of quantifiable temporal dimensions (i.e. lifespan, intensity, and burstiness). Second, we conduct a qualitative analysis to investigate whether news stories of different issue types have different narrative patterns, regardless of their specific topic. We use a corpus of 50,385 political news articles in Spanish from 2018 as a case study, and propose a novel system to aggregate the articles into stories. Our results show that stories belonging to different issue types do have distinguishing behaviours, especially along the intensity dimension. At the same time, the qualitative analysis indicates a tendency to associate narrative patterns to issue types. This analysis shows the potential of using news stories as research units to study framing strategies.

The stories narrated by mass media shape the public perception of reality. Previous studies suggest that the capacity of media to influence the salience of issues depends on four parameters, namely: (1) obtrusiveness; (2) duration; (3) abstractness; and (4) dramatism. Soroka (2002) built upon these parameters by developing an issue typology. Prominent issues affect a relevant number of people directly and are thought to leave little room for media impact on public opinion. Sensational issues are initiated by a dramatic event but have little observable impact on the everyday life of the majority of the population. Governmental issues are perceived as abstract or as not having direct effect on people's lives and do not offer dramatic or exciting components. Assuming the validity of Soroka's typology, we ask whether and how issue-types give rise to different life patterns of news stories. We thus identify three measurable dimensions: (a.) lifespan (i.e., the number of days over which articles of a news story are published); (b.) intensity (i.e., the number of articles per day per story); and (c.) burstiness (i.e., the speed with which stories go from emerging to their climax). We collected a corpus of 50,385 political articles from major Spanish newspapers in 2018. News stories were generated by aggregating articles with K-means clustering and cosine similarity, which resulted in 82 news stories. The results of our empirical analysis indicate that different issue-types exhibit different behaviors when represented as news stories. The contribution of this work is threefold: (i) it introduces a methodology to extract news stories from newspapers, (ii) it shows how the dynamics of news publishing can be used to study the narrative of news stories and (iii) it presents empirical data on the media’s narrative for different types of issues.