This paper presents a novel approach to classifying news articles by topic using only their URLs, addressing growing challenges in accessing article text due to paywalls and scraping restrictions. By fine-tuning a DistilBERT transformer model on URL data alone, I demonstrate topic classification performance that matches or exceeds traditional approaches requiring article text. Across three benchmark datasets spanning multiple languages and over 660,000 articles from more than 11,000 news domains, this URL-based topic classifier achieved superior F1 scores compared to both conventional machine learning methods and existing URL-based techniques. While this method requires more computational resources than simpler topic classification approaches, it dramatically reduces data collection requirements, offering researchers a practical alternative when text access is limited. These findings suggest that news article URLs contain richer semantic information than previously recognized, opening new possibilities for large-scale news content analysis in increasingly restrictive digital environments.

Recently, platforms have been increasingly deploying generative AI (GenAI) to summarize user-generated content (UGC) into AI-generated summaries (AIGS). However, the potential bias in AIGS and its impact on the public remain inadequately examined. We used Weibo, a leading social media in China, as a case to investigate these important questions, focusing on public sentiments. Specifically, we explored whether AIGS are biased in representing emotions in UGC and whether such representation influences subsequent public sentiment. We empirically identify two sources of bias in the algorithmic processes underlying the production of AIGS from UGC: the sampling process, in which GenAI selects a subset of UGC, and the summarizing process, in which the summary is generated from the sampled content. Comparing emotions in AIGS, sampled UGC, and all UGC, we found evidence of bias in both processes. In our case, GenAI tends to favor positive UGC during the sampling process and produces summaries that further amplify this positivity, leading to an over-representation of positive sentiments in AIGS. Additionally, we utilized a Difference-in-Differences (DiD) design to explore AIGS in public sentiment dynamics. Findings suggest that AIGS alone are insufficient to influence public sentiment significantly. Overall, this study provides important implications for deploying GenAI in public online discussions.

The use of online media has led to an increase in digital footprints of human behavior, which has resulted in a growing interest in the collection and analysis of such data in non-intrusive ways. One such approach is digital data donation, which involves requesting participants to share data that they have requested from digital platforms with researchers. While this approach promises to provide an unprecedented level of detail for computational communication research, it also raises concerns about the representativeness and validity of the data. This study investigates the issue of potential selection bias and non-response bias in data donation samples. It aims to identify subparts of the population that might be underrepresented (or even absent) in data donation sampling methods, which might lead to inaccurate and biased research conclusions. Utilizing a survey with a sample frame of 1178 and 289 participants from the Netherlands, the study investigates the relation between demographics, knowledge, privacy and trust factors, and donation behavior. The results show biases in donation behavior in terms of age and digital and algorithmic efficacy, while privacy and trust factors are not related to it. It suggests that individuals decide not to donate to academic research due to skills and not due to concerns or lack of trust. This offers possibilities for improvement in study design so that all willing individuals are equally able to participate.

This paper introduces grafzahl, an R package for fine-tuning Transformers for text data from within R. The package is used in this paper to reproduce the analyses in other papers. Very significant improvement in model accuracy over traditional machine learning approaches such as Convolutional Neural Network is observed.

Text-as-data methods have revolutionized the study of political behavior and communication, and the increasing availability of multilingual text collections promises exciting new applications of these methods in comparative research. To encourage researchers to seize these opportunities, we provide a guide to multilingual quantitative text analysis. Responding to the unique challenges research faces in multilingual analysis, we provide a systematic overview of multilingual text analysis methods developed for political and communication science research. To structure this overview, we distinguish between separate analysis, input alignment, and anchoring approaches to cross-lingual text analysis. We then compare these approaches’ resource intensiveness and discuss the strategies they offer for approaching measurement equivalence. We argue that to ensure valid measurement across languages and contexts, researchers should reflect on these aspects when choosing between approaches. We conclude with an outlook on future directions for method development and potential fields of applications. Overall, our contribution helps political and communication scientists to navigate the field of multilingual text analysis and gives impulses for their wider adoption and further development.

<div class="page" title="Page 1"> <div class="layoutArea"> <div class="column"> <div class="page" title="Page 1"> <div class="layoutArea"> <div class="column"> This study delved into cross-platform information flow and multilingual text analysis by examining social media posts on Weibo and Twitter in Chinese and English. We investigated public opinions about a violent restaurant attack in China that received widespread attention and validated three strategies of Bidirectional Encoder Representations from Transformers (BERT) to classify multilingual social media posts regarding their attitudes, targets, and frames. This study found that there was more criticism than support on Twitter than on Weibo when calling for social justice. When targeting the governments, Weibo users focused more on the local level, while Twitter users focused more on the state level. When framing their opinions, Weibo users focused more on gender violence, while Twitter users focused more on gang violence. These variations within social media posts across platforms were fundamentally influenced by the interruption of transnational information flow as a result of Chinese governance and censorship of the internet. Through the “porous censorship,” social media users’ autonomy and trust in the government played critical roles in the dynamics between online criticism and authoritarian responsiveness. </div> </div> </div> </div> </div> </div>

This paper introduces the 4CAT Capture and Analysis Toolkit (4CAT), an open-source Web-based research tool. 4CAT can capture data from a variety of online sources (including Twitter, Telegram, Reddit, 4chan, 8kun, BitChute, Douban and Parler) and analyze them through analytical processors. 4CAT seeks to make robust data capture and analysis available to researchers not familiar with computer programming, without ‘black-boxing’ the implemented research methods. Before outlining the practical use of 4CAT, we discuss three ‘affordances’ that inform its design: modularity, transparency, and traceability. 4CAT is modular because new data sources and analytical processors can be easily added and changed; transparent because it aims to render legible its inner workings; and traceable because of automatic and shareable documentation of intermediate analysis steps. We then show how 4CAT operationalizes these features through a description of its general setup and a short walkthrough. Finally, we discuss how 4CAT strives for an ‘ethics by design’ development philosophy that enables ethically sound data-driven research. 4CAT is then positioned as both an answer to and a further call for ‘tool criticism’ in computational social research.

A potentially powerful method of social-scientific data collection and investigation has been created by an unexpected institution: the law. Article 15 of the EU’s 2018 General Data Protection Regulation (GDPR) mandates that individuals have electronic access to a copy of their personal data, and all major digital platforms now comply with this law by providing users with “data download packages” (DDPs). Through voluntary donation of DDPs, all data collected by public and private entities during the course of citizens’ digital life can be obtained and analyzed to answer social-scientific questions – with consent. Thus, consented DDPs open the way for vast new research opportunities. However, while this entirely new method of data collection will undoubtedly gain popularity in the coming years, it also comes with its own questions of representativeness and measurement quality, which are often evaluated systematically by means of an error framework. Therefore, in this paper we provide a blueprint for digital trace data collection using DDPs, and devise a “total error framework” for such projects. Our error framework for digital trace data collection through data donation is intended to facilitate high quality social-scientific investigations using DDPs while critically reflecting its unique methodological challenges and sources of error. In addition, we provide a quality control checklist to guide researchers in leveraging the vast opportunities afforded by this new mode of investigation.

Ongoing research into how states coordinate foreign disinformation campaign has raised concerns over social media’s influence on democracies. One example is the spread of Russian disinformation in the 2016 US presidential election. Russia’s Internet Research Agency (IRA) Twitter accounts have been known to deliver messages with strategic attempts and political goals. We use publicly available IRA Twitter data created during and after the 2016 US election campaign (2016 and 2017) to examine the nature of strategic message features of foreign-sponsored online disinformation and their social media sharing. We use computational approaches to identify unique syntactic features of online disinformation tweets from IRA compared to American Twitter corpora, reflecting their functional and situational differences. More importantly, we examine what message features in IRA tweets across syntax, topic, and sentiment were associated with more sharing (retweets). Implications are discussed.