Multilingual Multifaceted Understanding of Online News in Terms of Genre, Framing, and Persuasion Techniques

Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics Volume 1: Long Papers, pages 3001–3022 July 9-14, 2023 ©2023 Association for Computational Linguistics Multilingual Multifaceted Understanding of Online News in Terms of Genre, Framing and Persuasion Techniques Jakub Piskorski1, Nicolas Stefanovitch2∗, Nikolaos Nikolaidis3, Giovanni Da San Martino4, Preslav Nakov5 1Institute of Computer Science, Polish Academy of Science, Poland jpiskorski@gmail.com 2European Commission Joint Research Centre, Italy nicolas.stefanovitch@ec.europa.eu 3Dept. of Informatics, Athens University of Economics and Business, Greece nnikon@aueb.gr 4Department of Mathematics, University of Padova, Italy dasan@math.unipd.it 5Mohamed bin Zayed University of Artificial Intelligence, UAE preslav.nakov@mbzuai.ac.ae Abstract We present a new multilingual multifacet dataset of news articles, each annotated for genre (objective news reporting vs. opinion vs. satire), framing (what key aspects are high- lighted), and persuasion techniques (logical fallacies, emotional appeals, ad hominem at- tacks, etc.). The persuasion techniques are an- notated at the span level, using a taxonomy of 23 fine-grained techniques grouped into 6 coarse categories. The dataset contains 1,612 news articles covering recent news on current topics of public interest in six European lan- guages (English, French, German, Italian, Pol- ish, and Russian), with more than 37k anno- tated spans of persuasion techniques. We de- scribe the dataset and the annotation process, and we report the evaluation results of mul- tilabel classification experiments using state- of-the-art multilingual transformers at differ- ent levels of granularity: token-level, sentence- level, paragraph-level, and document-level. 1 Introduction Internet has changed profoundly the information landscape by creating direct channels of commu- nication between information producers and con- sumers. At the same time, it has also increased the risk for readers to be exposed to disinformation (aka “fake news”), propaganda, manipulation, etc., which has grown into an infodemic (Alam et al., 2021). The consequences of this are very concrete, as swaying the hearts and the minds of a population also sways their choices, notably during elections. Therefore, online media analysis is important in order to understand the news ecosystem and the presented narratives around certain topics across countries, and to identify manipulation attempts and deceptive content, in order to provide citizens with a more transparent and comprehensible under- standing of the online news. ∗ The first and the second author have equally contributed to the work reported in this paper. Given the scale of the media landscape, media analysis needs automatic tools, which in turn need training data. With this in mind, we introduce a new dataset that covers several complementary aspects of the news: genre (objective news reporting vs. opinion vs. satire), framing (what key aspects are highlighted), and persuasion techniques (logical fallacies, emotional appeals, personal attacks, etc.). We collected news articles between 2020 and mid-2022, from sources ranging in the whole po- litical spectrum and revolving around widely dis- cussed topics such as COVID-19, climate change, abortion, migration, the Russo-Ukrainian war, and local elections. Our dataset is multilingual (English, French, German, Italian, Polish, and Russian), mul- tilabel, and covers complementary dimensions for better news understanding. Our taxonomy of per- suasion …

Content truncated to 500 words