Computational Communication Research - Volume 6, Issue 1, 2024

Times of crisis are usually associated with highly emotional experiences, which often result in emotionally charged communication. This is especially the case on social media. Identifying the emotional climate on social media is imperative in the context of crisis communication, e.g., in view of shaping crisis response strategies. However, the sheer volume of social media data often makes manual oversight impossible. In this paper, we therefore investigate how automatic methods for emotion detection can aid research on crisis communication and social media. Concretely, we investigate two Dutch emotion detection models (a transformer model and a classical machine learning model based on dictionaries) and apply them to Dutch tweets about four different crisis cases. First, we perform a validation study to assess the performance of these models in the domain of crisis-related tweets. Secondly, we propose a framework for monitoring the emotional climate on social media, and assess whether emotion detection models can be used to address the steps in the framework.

Citizen-generated populism is flourishing in the comments sections of online news. The factors that shape the extent of such populist communication from below are still under-researched. This study focuses on the COVID-19 crisis to examine how contextual and media-related factors are related to the extent of populism in comment sections on Facebook pages of news outlets from seven European countries (AT, DE, FR, IT, NL, SE and UK). Computational text analysis, machine translation and Bayesian multilevel regression were used to analyze digital trace data from 65,258 posts and 3.4 million comments published between February 2020 and June 2021. The computational measurements - multilingual dictionaries for posts and distributed dictionary representation to capture populism in comments - were rigorously validated. The results show that posts referring to the government, experts, COVID-19, and restrictions exhibit higher levels of populism in the comments sections. The stringency of containment policies was positively associated with populism in Germany, Austria, and the Netherlands when COVID-19 was mentioned. Lower levels of populism were observed for tabloid media and when news outlets engaged in visible moderation. The implications of these findings beyond the pandemic context and methodological challenges are discussed.

Understanding the decentralized formation of public opinion is increasingly important to communication research. Although many key determinants cannot be observed empirically, we argue they can be explored through theoretical modeling. Building on an existing agent-based model of opinion dynamics, our study introduces more complex, but theoretically interesting and realistic, agent behavior. We model distinct opinion tendencies which represent individuals’ diversity of belief, as well as external influences such as new information. Diversity increases the extremity of opinion in simulated consensus, radicalization and polarization. Simulation of new information demonstrates the ability of a minority group to shift majority opinion significantly in the long term, even with transient changes in behavior. Opposing minority groups do not counteract each other when their actions are delayed and may in fact amplify the original effect. We argue that modeling can help researchers and other stakeholders understand how these outcomes could arise in the real world, and thereby explore potential mitigations or exploitations.

Numerous studies document the gender gap in published articles in political science journals, observing systematic imbalances in the sub- mission pool which result in a distorted publication pattern. In this study we test some pathways that may explain the distorted submission pool: a) playing it safe due to the gender perception gap, and b) as a consequence of the Matilda effect setting a higher bar for methodological knowledge, focusing on papers using Computational Text Analysis methods. Papers using Computational Text Analysis Methods are more likely to be published in journals with a ‘masculinized’ perception gap. When women are aiming for these journals, they might ‘play it safe’ by conducting more validation checks than their male colleagues. More- over, embracing the Matilda effect – i.e. internalizing the systematic under-recognition of female scientists and mis-attribution of, especially methodological skills, to men – women scholars are more likely to indicate that a) there are important training needs in more areas; and b) they themselves need (further) training in computational methods and use these reasons not to publish papers employing these methods. We test these claims using a) a unique content analysis of research articles published in the top 20 journals in communication science, political science, sociology and psychology between 2016 and 2020, identifying all 854 articles that involved some form of quantitative textual analysis; and b) a pre-registered expert survey of all authors of quantitative text analytic research identified via said content analysis, which inquired about researchers’ considerations and concerns in the application of computational text analytic strategies

Computational methods are in full swing in communication science. Part of their promise is to make communication research more reproducible. However, how this plays out in practice has not been systematically studied. We verify the reproducibility of the entire cohort of 30 substantive and methods papers published in the journal Computational Communication Research (CCR), the official journal of the ICA Computational Methods Division with a focus on transparency and hence a high rate of voluntary Open Science participation in the field. Among these CCR papers, we are not able to verify the computational reproducibility of 16 papers as no data and/or code were shared. For the remaining 14 papers, we attempt to execute the code shared by the original authors in a standardized containerized computational environment. We encounter a variety of issues that preclude us from reproducing the original findings, where incomplete sharing of data or code is the most common issue. In the end, we could at least partially reproduce the findings in only 6 papers (20%). Based on our findings, we discuss strategies for researchers and the subfield to correct for this disheartening state of computational reproducibility.