Modeling Topic Networks on Highly Specialized Scientific Production Foci: A Case Study.

How does scientific knowledge about specialized research topics develop structurally? Existing literature on knowledge production networks shows the topology of networks in science, but this work often uses sampling frames (disciplines and sub-disciplines) that fail to capture the specific object of inquiry as a key data-generating feature of publication activity. Such work also uses tie indicators (collaborations or co-citations) that poorly approximate the interconnected knowledge embodied by research with ties between actors instead of related content. This paper (1) advances two methodological innovations that could benefit studies of scientific production, (2) demonstrates these approaches in a case study in the biochemical sciences, and (3) discusses the benefits and limitations of this approach. Methodologically, I first argue that network samples defined by shared research objects rather than disciplines, subfields, or common keywords can yield novel insight into how ideas develop in topic-specific networks and reveal both existing and potential opportunities for knowledge exchange across disciplines with different semantic norms. I also propose that structural topic modeling can be used to derive the latent content of journal articles addressing a particular research object, and that such an approach has unique advantages for capturing the broader knowledge structure on a specialized topic over collaboration, co-citation, and shared keyword methods. I then demonstrate the empirical viability of this pair of methodological innovations using research in both chemistry and biomedicine about a specific protein. Results show that topic-specific papers in the chemistry and biomedicine communities are structurally distinct, and that such structural differences are associated with differing research practices in the case study. I discuss these exploratory results and consider the limitations of the novel methods that produce them, then briefly elaborate on how specialized knowledge network dynamics can be studied and tested in future research. [ABSTRACT FROM AUTHOR]