ELTE Research Center for Computational Social Science - RC2S2

Gratulálunk az EKÖP (Egyetemi Kiválósági Ösztöndíj Program) 2026/27-es tanévi nyerteseinek! Kutatóközpontunkban érintett Buda Jakab (témája a parlamenti diskurzusban kormányváltások során bekövetkező változások feltárása gépi tanulás alapú kvantitatív adatelemzéssel) és Vajda Dorina (a mesterséges intelligencia munkaerőpiaci hatásainak reprezentációja az online sajtóban és annak természetesnyelv-feldolgozás alapú mérése). Mindketten a kutatócsoport keretein belül végzik majd munkájukat. Sok örömet és sikert kívánunk a kutatásokhoz!

A konferencián kutatóközpontunk három tagja (Buda Jakab, Rakovics Zsófia és Vajda Dorina) mutatja be kutatását.

EKÖP zárókonferencia TáTK-s előadókkal (2026) Az Egyetemi Kutatói Ösztöndíj Program ELTE-s ösztöndíjasai számolnak be az év során elért tudományos eredményeiről.

Kutatócsoportunk tagjának, S*k Domonkosnak az írása a klímaválság társadalmi kontextusáról

A hőkupola árnyékában Hasonlóképpen a világjárványhoz, most is olyan társadalmi válság elé nézünk, ami minden nehézsége ellenére kivételes lehetőségeket is magában rejt. S*k Domonkos írása.

Our new paper in Mass Communication and Society (D1).

The research project titled “Representations of Migration in a Transforming Europe,” funded by the CELSA Research Fund, took place between 2023 and 2025 as a collaboration between KU Leuven and ELTE RC2S2 (principal investigators: Leen d’Haenens (KU Leuven) and Ildikó Barna (ELTE)). This recently published paper is one of the outcomes of that research.
The paper compares media framing of the 2015 Syrian and 2022 Ukrainian refugee crises across two contrasting media systems: Belgium's Dutch-speaking Flanders and Hungary. The former represents a democratic corporatist model, while the latter aligns more closely with a polarized pluralist system. Drawing on media systems and framing theory, we examined how institutional structures shape the representation of refugees. Using structural topic modeling on more than 85,000 newspaper articles, we identified dominant topics and trace framing differences across time and national contexts. We found that both media systems portrayed Ukrainian refugees more favorably than Syrian refugees, but the degree and nature of this shift varied considerably. Flemish media demonstrated more consistent and pluralist coverage across crises, whereas Hungarian media exhibited greater framing volatility and closer alignment with state narratives. This study contributes to debates on the democratic role of independent media and demonstrates the value of combining computational and interpretive methods in comparative framing analysis.

Free eprint link, it can be used up to 50 times:
https://www.tandfonline.com/eprint/4AMVXMGZJUXGCKJRTTVR/full?target=10.1080/15205436.2026.2698720

Abstract with open access at research institutes and universities in Hungary:
https://www.tandfonline.com/.../10.../15205436.2026.2698720

Another threat to sound scientific methods: LLMs on the hunt for significance

(PDF) LLMs p-hack LMMs PDF | My recent work has shown that current AI systems help users carry out various types of questionable research practices (QRPs). Building upon the... | Find, read and cite all the research you need on ResearchGate

We met Viktor Berger at the Hungarian Sociological Association’s conference in Pécs, where both he and we (Renáta Németh, Miklós Szabó) gave presentations on LLMs, and when it turned out that he, too, had been interested in sci-fi as a teenager, we decided to collaborate on a research project - which became the topic of anthropomorphism and LLM use. The results of our joint research have just been published.

This study utilizes a qualitative methodology informed by phenomenology to examine student–machine interactions in higher education, investigating whether students attribute human-like qualities to the machine. Drawing on Edmund Husserl’s concepts of active and passive consciousness, alongside recent phenomenological and anthropological work on AI, we analyze a substantial body of data tracking student–AI tutor interactions across various disciplines using a sociolinguistic approach. Although this humanization may facilitate learning, it raises critical ethical concerns regarding excessive trust - at the end of our paper, we offer several recommendations for future developments.

https://www.sciencedirect.com/science/article/pii/S2772503026000599

Kutatócsoportunk tagjai - Buda Jakab és Németh Renáta új publikációja. A tanulmány a Magyar Országgyűlésben 1998 és 2018 között elhangzott beszédeket elemezve a szöveges adatokon végzett felügyelt és felügyelet nélküli gépi tanulás kombinációjának alkalmazási lehetőségeit mutatja be. A polarizáció mértékét n-gram alapú XGBoost klasszifikációs modellel mérték, amely a két domináns párt (Fidesz és MSZP) képviselőinek beszédeit különbözteti meg. Ezt kiegészítve 16 topikból álló strukturális topikmodellt alkalmaztak a polarizáció tartalmi csomópontjainak feltárására. Az egyes topikok polarizáltságát a modell biztossága alapján határozták meg, és ez alapján azonosították a polarizáltabb és kevésbé polarizált témákat, majd ezeket mélyebben is megvizsgálták. A szerzők tudomása szerint ilyen vagy hasonló módszerkombinációval még nem kísérleteztek korábban, de eredményeik alapján érdemes lehet más esetekben is használni, amikor csoportok nyelvhasználati különbségének értelmező feltárása a kutatási cél.

www.ksh.hu

Traditionally, awards for the best thesis in each program are presented at the end-of-year graduation ceremony. This year, in the MSc in Survey Statistics and Data Analytics, András Richárd Wernigg's thesis won this title based on the votes of the program's instructors, in a very close competition, as he had excellent competitors. His thesis is a perfect example of how statistics can be of assistance to related disciplines.

András Richárd Wernigg - Application of Monte Carlo Simulation in the Analysis of Healthcare Quality and Risks: A Complex Decision-Support Model Using Hernia Surgery as an Example – ELTE Research Center for Computational Social Science András Richárd Wernigg (LinkedIn) This research examines the quality and accessibility of hernia care in Hungary using stochastic Monte Carlo simulation. The model, based on 2022 baseline data, demonstrates the dangers of deterministic capacity planning and the “error of averages.” The 10-year...

Our students have the floor - Theses in CSS, part 7.
An analysis of political polarization in Hungarian online media through the lens of mentions of individuals.
In his thesis, János Sebestyén Pap: (supervisor: Zsófia Rakovics, Survey Statistics and Data Analytics MSc) examines political polarization in Hungarian online news media through patterns of mentions of individuals’ names. The research is based on the hypothesis that polarization manifests not only in opinions but also in which public figures the media highlight over others, and what structural patterns and groupings can be observed among them. The empirical analysis is based on a text corpus comprising articles published between 1998 and 2022. The study combines natural language processing and network analysis methods, using named entity recognition to identify individuals and then constructing news site–person type-pair networks. The study examines the differentiation of news sites based on co-occurrence within and between groups, as well as a polarization index calculated from these data. The results indicate significant polarization following 2017, which is consistent with previous research on the political fragmentation of the Hungarian media system. This study contributes to the network-based measurement of media polarization; however, its limitations include the accuracy of entity recognition and the representativeness of the data.

Details in Hungarian:
https://rc2s2.elte.hu/thesis/pap-janos-sebestyen-a-magyar-online-media-politikai-polarizaciojanak-vizsgalata-a-szemelyek-emlitesen-keresztul/
Details in English:
https://rc2s2.elte.hu/en/thesis/janos-sebestyen-pap-an-analysis-of-political-polarization-in-hungarian-online-media-through-the-lens-of-mentions-of-individuals/

In our new study, De-Democratizing the Public Sphere – A Structural Topic Model Analysis of Hungarian Parliamentary Discourses Between 1998-2020 (Problems of post-communism, Q1), we examined more than two decades of parliamentary debates through the lens of Habermas's theory of communicative rationality. Using structural topic modeling, we identified long-term changes in the themes and communicative patterns of Hungarian parliamentary discourse. Together, the trends point to a weakening of parliament's capacity to provide deliberative control over nepotistic and corrupt systems.
Since the 2026 election, however, we know that democratic backsliding was not the end of the story. The sudden collapse of the Orbán regime demonstrated that even seemingly stable populist systems can become vulnerable when they overstretch the discursive frames that sustain them.
Yet parliament itself did not play a central role in this turning point. On the contrary, our research suggests that it gradually lost relevance within the Hungarian public sphere throughout the period. The key transformations occurred outside parliament—in the broader media and public communication environment.
This is why our recent research has shifted its focus to the changing public sphere. In upcoming publications, we explore how a post-truth illiberal media sphere emerged, evolved, and ultimately collapsed under the weight of its own internal contradictions.
https://www.tandfonline.com/doi/full/10.1080/10758216.2026.2673869