Jinmo Kim's Dissertation Defense

Research funding shapes the conditions under which scientific knowledge is produced, yet large-scale analysis of funding systems depends on bibliographic metadata in which funding organization names are often inconsistent and ambiguous. This dissertation develops and applies computational approaches to construct reliable organization-level funding data and examine the organizational and sectoral structure of research funding in Library and Information Science (LIS), using Biodiversity Conservation (BC) as a comparative domain. The dissertation consists of three connected empirical studies. First, an organization name disambiguation (OND) framework combines manual annotation, traditional machine learning, neural models, fine-tuned Transformers, prompt-based large language models, and hierarchical clustering. In-domain learning substantially outperformed cross-domain transfer learning, with BERT providing the strongest pairwise and clustering performance, demonstrating the importance of domain alignment and consistent entity definitions. Second, the disambiguated data are used to examine funder dynamics and co-funding networks. Funding participation and co-funding increased in LIS, but publication-level growth did not correspond to comparable increases in organization-level network connectivity. Comparison with BC revealed different funding trajectories, showing that co-funding prevalence and network structural change represent related but distinct dimensions of funding dynamics. Third, funding organizations are classified as government or non-government to examine sectoral participation and co-funding. LIS showed simultaneous growth in government-only, non-government-only, and cross-sector funding rather than simple substitution between sectors. Government and non-government participation remained positively associated, although this relationship weakened over time. BC exhibited a comparatively stable and highly cross-sectoral funding structure. The comparison also highlights the need for cautious cross-domain interpretation when bibliographic sources and observation periods differ substantially. Together, the findings establish a sequential framework that moves from metadata standardization to organizational and sectoral analysis. The dissertation demonstrates that reliable funder identification is a prerequisite for valid downstream analysis and that research funding is best understood as a dynamic system of interacting organizations and sectors.

Questions? Contact Jinmo Kim.