Dávid Angyalffy (LinkedIn, GitHub, E-mail)
The polarization of the Hungarian online media landscape and the quantitative assessment of changes in editorial policy are current issues in digital journalism research. This study analyzes whether a change in the language of the Index.hu portal can be detected following the 2020 change in ownership and editorial leadership. The analysis employs a proxy-based approach. In this process, machine learning models are trained on articles from two news portals representing different editorial policies—HVG and Origo—and these models are then applied to Index articles. The analysis compares the performance of three model architectures—logistic regression, XGBoost, and BiLSTM with Hungarian fastText embeddings—supplemented by SHAP- and LIME-based interpretability analyses. Statistical validation was performed using the Mann-Whitney U test and Cohen’s d effect size. The results suggest that the average P(Origo) value of Index articles from 2021 is consistently higher than that of articles from 2019 for all three models, indicating a systematic shift toward the Origo stylistic pole.