Eine Übersicht aller Sessions/Sitzungen dieser Veranstaltung. Bitte wählen Sie einen Ort oder ein Datum aus, um nur die betreffenden Sitzungen anzuzeigen. Wählen Sie eine Sitzung aus, um zur Detailanzeige zu gelangen.
Communication science aims to gain insight into the nature of media content and its impact on recipients. Regarding the former, communication science has developed the methodology of quantitative content analysis. About the latter, communication scientists frequently employ experimental designs. To test the effect of specific message features, stimuli should vary only in the characteristic(s) of interest. Manipulating stimuli, however, carries the risk of accidentally changing multiple characteristics simultaneously. In such cases, researchers incorrectly attribute the differences in outcomes between experimental groups to the content feature of interest. Our paper combines these two perspectives fruitfully by using the advantages of artificial intelligence. We create prompts for Large Language Models (LLMs) based on data from a content analysis, which instruct the LLMs to create an unlimited number of texts with specific content features but varying wording, which helps to compensate for the risk of confounding.
How to link far-right user identities across social media platforms: A computational exploration
Azade Kakavand1, Frederik Møller Henriksen2, Marvin Stecker1, Ahrabhi Kathirgamalingam1, Alexander Dalheimer1, Annie Waldherr1
While influential actors use multiple social media platforms, most research is focused on single-platform communication. Borrowing from the computer science literature on User Identity Linkage, we aim to evaluate different techniques to match user identities of public pages on Facebook to the corresponding X accounts. To do so, we employ methods that are already known to communication scientists, such as fuzzy matching and sentence embeddings, and compare the results to a hand-coded gold standard. We focus on the far right as a relevant use case since they are a heterogeneous group that heavily uses social media for communicating. Ultimately, the findings should be applicable beyond the far right to other (political) influential actors as well and help communication scientists to analyze actors’ social media performances as a whole.
How we talk about liberal latte drinkers: Measuring Group Politicisation in Political Texts
Marvin Stecker1,2, Fabienne Lind1, Hajo G. Boomgaarden1, Markus Wagner2
1Institut für Publizistik- und Kommunikationswissenschaft, Universität Wien, Österreich; 2Institut für Staatswissenschaft, Universität Wien, Österreich
Group based conflict is an inevitable feature of democracy. Yet, we are missing empirical methods that would allow us to trace the role that elite actors have in promoting or influencing group contestation. We outline a framework to trace the politicisation and polarisation around social groups in political texts.
First, we utilise a fine-tuned transformer model to track the salience of different social groups, inductively extracting mentions of them at the word level. Secondly, we estimate the values attached to social group references using static word embeddings.
We will utilise data from political parties as well as newspaper reporting, as these are two important groups of actors shaping national debates. We cover Austria and the UK over a ten-year time period, providing both descriptive and time-series analysis of the contestation of social groups.
