Mitchell Centre Seminar Series

About

Mark Wittek CEU Network Ecologies of Cultural Production: Evidence from Jazz and Science Scholars have studied the social organization of scientific and cultural fields extensively through a network-analytical lens. Several of these accounts stress that the formation of networks depends on contextual characteristics, such as institutional, historical, and material conditions that shape how people collaborate, exchange ideas, and accumulate recognition. In this talk, I propose an extension of the network ecology framework to study the link between network mechanisms—such as preferential attachment, homophily, and influence—and contextual characteristics in scientific and cultural fields. Testing network-ecological propositions at scale requires models that capture whether identical mechanisms operate differently across communities, genres, or epistemic cultures, and relate those differences to characteristics of these settings, for instance, their size, age, and demographic composition. Two challenges arise: interpreting process-generated data and the limits of statistical network models for handling large networks. During the talk, I will showcase two lines of research that address these challenges and provide preliminary large-scale evidence for contextual variation in network mechanisms across cultural and scientific fields. The first example study, on artistic influence among jazz trumpet players, combines an expert-curated genealogy with musicians' discographies and audio embeddings of their recordings, showing that influence is not reducible to sounding alike: shared collaborators matter, and highly similar musicians are less likely to be linked by influence. The study also illustrates that interpretive work is needed to turn process-generated traces into valid measures of cultural content. For example, we find that audio embeddings respond strongly to the recording technologies used in different eras of jazz music. The second study introduces contextual relational hyperevent models (C-RHEMs) to overcome limitations related to the degeneracy of statistical network models in large datasets. We apply C-RHEMs to over 400,000 publications in oncology, which is a scientific field large and internally heterogeneous enough to make contextual variation visible. The study shows that across 32 topics, preferential attachment, gender homophily, and self-citation vary substantially, and contextual characteristics partly explain this variation: older topics show greater inequality, more consolidated self-citation, and stronger gender homophily. Together, the two cases sketch a scalable framework for examining the intersection of culture and networks, extending the empirical reach of network ecology to science, music, and other domains of cultural production.