What is it about computational communication science?

In this episode of the #aBitOfCCS Podcast, Jana Bernhard-Harrer talks to Miriam Milzner and Daniel Thiele from the Weizenbaum Institute’s research group on the Dynamics of Digital Mobilisation about detecting coordinated online manipulation in the era of generative AI.They introduce coorsim, an open-source R package that identifies semantically similar coordinated posting—addressing a key limitation of traditional tools that rely on identical or near-identical text. Evaluated across 15 international influence operations, coorsim demonstrates how embedding-based similarity and coordination-sensitive clustering can uncover sophisticated campaigns, even when content is linguistically diverse. The episode also explores an example of coordinated activity during the climate summits COP26 and COP27, drawing on over 5.8 million tweets. Miriam and Daniel reflect on how coordinated campaigns shape climate debates—and what this means for research on disinformation in the age of LLMs.GitHub Repository: https://github.com/thieled/coorsimEmailMiriam: miriam.milzner@fu-berlin.deEmail Daniel: daniel.thiele@fu-berlin.de

In this episode, Professor Mario Haim from LMU Munich breaks down what it really means for research to be robust — covering key concepts like generalizability, validity, reliability, reproducibility, and replicability. Mario explains how these ideas connect and why they matter, especially when studying opinionated communication with computational methods. He shares practical insights on common pitfalls, specific challenges in this field, and concrete steps researchers can take to strengthen their studies — from better documentation to transparent workflows and careful methodological choices. Whether you’re designing a new project or refining your research practices, this conversation offers valuable guidance for ensuring your work stands on solid ground.

In this episode of aBitOfCCS, Keonhi Son (Mannheim Centre for European Social Research) discusses her study on how women were talked about in the Weimar Republic’s parliament between 1919 and 1932. Using quantitative text analysis and semantic network methods, Keonhi examines how terms such as woman, mother, homemaker, and (female) worker were used in Reichstag debates from 1920 to 1932 — and how these meanings varied by political party, ideology, and gender of the speaker. The conversation sheds light on how early 20th-century German politics framed women’s roles and how those discourses both reflected and shaped broader social change.📧 Questions? Contact Keonhi at son@uni-mannheim.de

In this episode, Dr. Aleksandar Tomašević from the University of Novi Sad takes us beyond text-based analysis to explore how emotions expressed in videos can be measured and understood. Aleksandar explains why studying non-verbal cues—especially facial expressions—is becoming crucial for understanding political communication online. He walks us through different methods for detecting these expressions, highlighting how machine learning and deep learning techniques enable computational analysis of emotions. Aleksandar also discusses the accuracy of machine-based emotion detection compared to human judgment and shares fascinating findings from his research on political leaders’ emotional expressions in video content. This conversation reveals how emotion analysis opens new doors in communication research.

In this episode, we’re joined by Dr. Johannes Gruber from Vrije Universiteit Amsterdam to unpack the world of language models. Johannes explains what language models really are and how they shape how we interact with information — from powering everyday chatbots like ChatGPT to supporting advanced research. We break down how these systems work behind the scenes, what they’re great at, and where we need to be cautious. Johannes also shares insights from his recent research on the feedback loops between language models, citizens’ beliefs, and democracy. It’s a closer look at why understanding both the potential and the limits of language models is so important for opinion research today.

Tune in to the #aBitOfCCS Podcast as we explore the computational workflow behind digitizing a historical society magazine. Christian Lendl joins us to discuss his paper Digitizing the Aristocratic Elite: Computational Challenges and Methods in Processing the Wiener Salonblatt (1870–1938). The episode highlights how AI-driven workflows can open new possibilities for digital humanities research.Reach out to Christian at christian@lendl.pro

In this episode, we’re joined by Prof. Eetu Mäkelä from the University of Helsinki to break down the world of word embeddings. Eetu explains what word embeddings are in simple terms, how they fit into the bigger picture of language models, and why they’re so powerful for exploring relationships in language — from the famous King–Queen example to applications in studying opinions. We look at how researchers can work with pre-trained embeddings or build their own, and how these tools open new ways to analyse language and meaning at scale. Eetu also shares where research on word embeddings is headed next and why they remain central to the evolving field of opinionated communication.

Tune in to the #aBitOfCCS Podcast as we dig into the growing tension between performance and sustainability in computational text analysis. Sean Palicki, a researcher at TUM, joins us to discuss his recent paper Don’t Look Up: Evaluating the Tradeoff between Performance and Sustainability of LLMs for Text Analysis.In this episode, we explore how large language models (LLMs) compare to lighter methods such as dictionaries and task-specific classifiers when applied to sentiment analysis, classification, and named entity recognition in political texts. We talk about the environmental costs of relying on large models, why bigger doesn’t always mean better for text analysis, and how introducing a CO₂-adjusted F1 score can help balance accuracy with sustainability.The conversation highlights a “right-fit” approach to model selection—choosing tools that are not only effective but also environmentally responsible.Reach out to Sean at sean.palicki@tum.de and find his website here: https://sean.web-of-us.com/ 

In this episode, we’re joined by Prof. Damian Trilling from Vrije Universiteit Amsterdam, who opens the door to the world of machine learning for opinion research. Damian explains how citizens consume and share news today — and how machine learning helps us make sense of these patterns at scale. We unpack the difference between supervised and unsupervised machine learning and explore how blending both can strengthen research projects. Damian also shares why these methods hold so much promise for the future of studying opinionated communication and news use in the digital age.

Tune into the #aBitOfCCS Podcast as we explore how large language models classify online political speech about sexism. Aditi Dutta, a doctoral researcher at the University of Exeter, joins us to discuss her study on how automated moderation systems often misclassify anti-sexist speech as harmful—raising important questions about fairness, resistance, and digital democracy.CONTENT WARNING: This episode includes discussions and examples of sexist language online, which may be offensive or upsetting to some listeners.Read the paper here: https://arxiv.org/abs/2508.11434v1Reach out to Aditi at ad882@exeter.ac.uk for more insights into her research.