The science of science is an emerging transdisciplinary field that leverages large-scale data to uncover patterns in scientific activity, informing both policy decisions and the work of individual researchers. A central challenge in the science of science is mapping the relationships between scientific publications. Traditional approaches rely on metadata such as citations or shared keywords, which capture only limited aspects of these links. Recent advances in large language models (LLMs), however, open new possibilities for extracting richer content-based information, including causal claims, methodological details, and sample characteristics. This project therefore explores and evaluates methods for information extraction and concept normalization in the social sciences, comparing them in terms of accuracy and efficiency. The results will provide valuable tools and benchmarks for researchers seeking to analyze and connect scientific knowledge more effectively.