Computational Communication Research - Current Issue

The introduction of Transformers, neural networks employing self-attention mechanisms, revolutionized Natural Language Processing, handling long-range dependencies and capturing context effectively. Models like BERT and GPT, trained on massive text data, are at the forefront of Large Language Models and have found widespread use in text classification. Despite their benchmark performance, real-world applications pose challenges, including the requirement for substantial labeled data and class balance. Few-shot learning approaches, like the Recognizing Textual Entailment framework, have emerged to address these issues. RTE identifies relationships between a text T and a hypothesis H. T entails H if the meaning of H, as interpreted in the context of T, can be inferred from the meaning of T. This study explores an RTE- based framework for classifying vaccine-related news headlines with only 751 labeled data points distributed unevenly across 10 classes. The study evaluates eight models and procedures. The results highlight that deep transfer learning, combining language and task knowledge, like Transformers and RTE, enables the development of text classification models with superior performance, effectively addressing data scarcity and class imbalance. This approach provides a valuable protocol for creating new text classification models and delivers an advanced automated model for classifying vaccine- related content.

During the COVID-19 pandemic, wearing a face mask became politicized in the United States, with politicians and reporters employing competing public safety and civil liberties frames in discussions of masking. In this research, we argue that political candidates’ decisions to speak about and depict mask-wearing in their political advertising were strategic, depending on both the candidate’s party and gender. We examine political ads run on Facebook and on television by federal candidates during the 2020 U.S. campaigns. We use Amazon’s deep learning algorithms for PPE (personal protective equipment) detection. We extract the text and audio of each ad to identify mentions of masks and use an à la Carte embedding regression model to understand how the usage of the term mask differs across covariates. We find that images of masks are much more common than mentions of masks, that there are significant partisan, but not gender, differences in the use of masks, and that there are both partisan and gender differences in the way that candidates speak about masking. This research demonstrates the utility of a novel approach to collecting data. It also suggests that public health measures can become partisan in a campaign environment, with the potential to polarize both the views and behaviors of Democrats and Republicans.

ammico (AI-based Media and Misinformation Content Analysis Tool) is a publicly available software package written in Python 3, whose purpose is the simultaneous evaluation of the text and graphical content of image files. After describing the software features, we provide an assessment of its performance using a multi-country, multi-language data set containing COVID-19 social media disinformation posts. We conclude by highlighting the tool’s advantages for communication research.

As media platforms continue to develop content moderation policies, alternative platforms have emerged as safe havens for deplatformed content. As these alternatives to major media platforms grow, the importance of understanding their role in the media ecosystem grows too. In this paper, we perform a longitudinal study of the content dynamics of one such alternative media platform, BitChute. BitChute is an alternative video-hosting site similar to YouTube. We first theorize what technological affordances may drive the supply and demand of content on BitChute. We then test those theories through an analysis of 6,363,596 videos from 82,162 channels, which were viewed 2,868,117,905 times, over 54 months. We find that BitChute’s minimal content moderation drives much of the content supply and demand. Videos which were more offensive, certain, and covered commonly deplatformed topics were most popular. In particular, we find that BitChute fills a demand gap created by moderation policies on major media platforms around COVID-19 and - to a lesser extent - elections fraud. The most popular videos on the platform were re-uploaded videos that were banned by YouTube and Facebook. As a whole, our results suggest that BitChute’s current role is less as a town square and more as a backup for deplatformed video content.

Misinformation has to be uncovered before it can be used for research purposes. This is a resource intensive process, which is why fact-checks have been a popular data source. They have been used directly as proxy for misinformation and indirectly to identify its sources and analyze its content and spread. However, there is little research on the limitations of fact-checks as a data source. Are there patterns in their topics that might lead to biased research results? How does the fact-checkers’ choice of targeting certain actors and social media platforms influence their article’s content? The study provides answers to these questions. It analyzes fact-checks from four German outlets between 2019 and 2023. The study finds that certain topics appear continuously, while for others coverage is event- driven. Furthermore, political actors are covered only to a small extent and even less when they are the originators of misinformation. Finally, fact-checks focus strongly on misinformation on Facebook and the findings indicate that the topic distribution of fact-checks might be different if other platforms were focused. The article discusses the findings with respect to limitations of fact-checks as a data source and concludes with practical recommendations for future research.