Abstract While considerable attention has been devoted to the comparison of media systems and structures, the analysis of actual media content across nations has received comparatively less attention and has relied largely on manual methods, particularly content analysis and discourse analysis. Topic modelling, which is increasingly utilised to analyse large-scale textual datasets, has emerged as a promising tool for comparative content analysis. However, it is often criticised for being overly descriptive, lacking theoretical integration, and facing challenges related to reliability, validity, and multilingual inclusivity. This paper examines BERTopic, a transformer-based topic modelling method, through a multilingual case study of climate change coverage across 40 newspapers in 13 countries. Our findings suggest that topic modelling is a valuable tool for presenting and visualising descriptive patterns, particularly when integrated with classical methods such as content analysis. It also offers the potential for greater cost-efficiency and scalability, which is particularly valuable in research settings characterised by unequal access to funding and high-performance computing infrastructure, including many parts of the Global South. However, we contend that topic modelling should be viewed as a starting point rather than a comprehensive solution. For meaningful contributions, it must be tied to broader theoretical frameworks and consider contextual factors, such as journalistic cultures and socio-political conditions, rather than merely presenting descriptive results in isolation as has been the case in many studies within the field. This study highlights the importance of combining computational tools with theoretical enquiry to enhance analytical depth in comparative media analysis and similar research in the social sciences.
Wang et al. (Fri,) studied this question.