GREENER: Graph Neural Networks for News Media Profiling

Proceedings of the 2022 Conference on Empirical Methods in Natural Language Processing, pages 7470 - 7480 December 7-11, 2022 ©2022 Association for Computational Linguistics GREENER: Graph Neural Networks for News Media Profiling Panayot Panayotov Sofia University p.d.panayotov09@gmail.com Utsav Shukla TIET Husrev T. Sencar QCRI, HBKU hsencar@hbku.edu.qa Mohamed Nabeel QCRI, HBKU mnabeel@hbku.edu.qa Preslav Nakov Mohamed bin Zayed University of Artificial Intelligence preslav.nakov@mbzuai.ac.ae Abstract We study the problem of profiling news media on the Web with respect to their factuality of reporting and bias. This is an important but under-studied problem related to disinforma- tion and “fake news” detection, but it addresses the issue at a coarser granularity compared to looking at an individual article or an individ- ual claim. This is useful as it allows to profile entire media outlets in advance. Unlike pre- vious work, which has focused primarily on text (e.g., on the articles published by the tar- get website, or on the textual description in their social media profiles or in Wikipedia), here we focus on modeling the similarity be- tween media outlets based on the overlap of their audience. This is motivated by homophily considerations, i.e., the tendency of people to have connections to people with similar inter- ests, which we extend to media, hypothesizing that similar types of media would be read by similar kinds of users. In particular, we propose GREENER (GRaph nEural nEtwork for News mEdia pRofiling), a model that builds a graph of inter-media connections based on their audi- ence overlap, and then uses graph neural net- works to represent each medium. We find that such representations are quite useful for predict- ing the factuality and the bias of news media outlets, yielding improvements over state-of- the-art results reported on two datasets. When augmented with conventionally used represen- tations obtained from news articles, Twitter, YouTube, Facebook, and Wikipedia, we im- prove over previous work by 2.5-27 macro-F1 points absolute for the two tasks and datasets. 1 Introduction The problem of news media profiling with respect to their factuality of reporting and political bias is important but under-studied. It is related to dis- information and “fake news” detection, but it is of different granularity compared to looking at an individual article or at an individual claim. This kind of profiling can be done by profes- sional fact-checkers, who inspect the articles and the multimedia material published by the target news outlet. However, doing this automatically while solely relying on text features is a very chal- lenging task as previous work has shown (Baly et al., 2018, 2020; Stefanov et al., 2020). It gets even more complicated when considering news sources where only limited number of examples is available for evaluation. Therefore, not only is there a need to characterize news media more thor- oughly, but there is also a need to be able to do so in a predictive fashion using limited information. A crucial consideration is the need to comple- ment the textual representation with other elements of a news medium that may …

Content truncated to 500 words