How Cities Can Measure Monument Disputes Using AI and News Data
Researchers in Romania developed a computational method to track how public monuments fuel ethnic tensions by analyzing decades of newspaper coverage with machine learning. The technique offers cities and policymakers a scalable way to monitor cultural flashpoints and anticipate conflicts before they escalate into political crises.
Originaltitel: Computational Analysis of Contested Monuments and Collective Memory in a Multiethnic City
ABSTRACT This study analyses how four public monuments in the centre of Cluj‐Napoca (the Matthias Corvinus Statuary Ensemble, Avram Iancu, Baba Novac and the Transylvanian School Statuary Ensemble) reflect Romanian‐Hungarian relations, urban governance practices and how collective memory is negotiated in post‐socialist public space. The research combines a qualitative survey of local and regional newspapers (1990–2025), field observation and GIS analysis, with a multi‐stage computational pipeline for event classification and sentiment analysis. The media corpus was coded in the CAMEO ontology and assessed by a multilingual natural language inference model (Zero‐Shot), complemented by an event intensity score (Goldstein) and two sentiment analysis branches: a Hugging Face (HF) and Valence Aware Dictionary and sEntiment Reasoner (VADER) transformative model, contextually adapted and lexically extended for the language of heritage‐related news. The results show that the Zero‐Shot scores remain low and relatively uniform, indicating a predominantly descriptive media discourse. The Goldstein scores highlight tense episodes for the statues of Avram Iancu and Baba Novac and high volatility for Matthias Corvinus, while the HF and, to a lesser extent, VADER, indicate a clear intensification of the emotional tone after 2022, especially for the Transylvanian School and Matthias Corvinus. The differences between the monuments are less pronounced than the differences between the analytical tools, but all series converge towards a recent increase in discursive intensity. The qualitative analysis outlines three main registers of contestation: (i) the spatial performance of monuments in central squares as spaces of urban power; (ii) disputes over inscriptions and textual formulations and; (iii) public trust in the administrative management of heritage. The conclusions suggest a shift from a strictly custodial conservation logic to a communicative governance of urban memory, based on procedural transparency, public digital registers, multilingual inscriptions and the systematic use of computational tools in planning and evaluating heritage policies.