Network Analysis of Co-authorship and Research Directions in Thermoelectrics

This paper presents a methodology for constructing an extended co-authorship network in the field of thermoelectrics by integrating direct co-authorship links with thematic proximity derived from researchers’ Google Scholar profiles. The network is built through targeted probing using relevant keywords (e.g., "thermoelectrics"), and link weights are determined by both the number of joint publications and shared research interests. This approach enables the identification of not only direct collaborators but also potential interdisciplinary partners, revealing the latent structure of the scientific community. Additionally, a descriptor network is constructed based on the co-occurrence of keywords across author profiles, forming an ontological map of the field. Thematic clusters are identified using the Louvain algorithm, representing core research directions. For the first time in this context, Large Language Models (LLMs) are employed to interpret cluster content by generating meaningful, human-readable labels from lists of descriptors. This allows for fast, objective, and scalable identification of scientific trends without relying on expert annotation. The analysis shows that the extended network exhibits higher density than the traditional co-authorship network, highlighting the significance of thematic connections. Centrality measures (degree and betweenness) help identify key contributors and structural bridges within the field. The proposed approach supports the analysis of scientific communities, detection of research schools, and collaboration forecasting