Mitchell Centre Seminar Series

About

Jana Diesner Technical University of Munich What we know and wrongly assume about the how and why of link formation in collaboration networks Large-scale bibliometric analyses allow us to validate classic theories of link formation in contemporary settings. I show how measurement choices can lead to erroneous theorizing about network formation principles, and how to correct for that. An alternative to bibliometrics as well as elicitation as methods for network data construction is relation extraction, i.e., locating and categorizing instances of nodes and edges from unstructured, natural language text data, and representing the extracted data in relational form, e.g., to enable subsequent network analyses. I am discussing our findings for using LLMs for this task and compare LLM performance to alternative relation extraction methods. I also present our findings about the ability of LLMs to capture the quality of social relationships and personal values from long-form text data.