Mapping the Physics Research Space: a Machine Learning Approach

Scientific research is a complex endeavor and the right mix of skills and competences is necessary to effectively produce new knowledge. In this work we describe how we can use an unsupervised machine learning algorithm to obtain a vector representation of scientific topics which in turn allows us to create a mapping of the research space in Physics. The research space is then used to explain the evolution of scientific expertise of urban areas over time and across scientific fields. In particular, we show that the breadth of expertise of researchers in specific sub-fields of Physics can be used to construct a science map by embedding research fields into an $N$-dimensional space.
Lastly, we demonstrate how knowing the research fingerprint of an urban area along with the overall configuration of the research space can be used to predict the evolution over time of the relative comparative advantage of a region. From a practical standpoint, our results can be used to inform policy makers and guide their decision making process when considering how to steer investments to alter or reinforce existing location-specific scientific competences and research trajectories. In addition, our methodology can be readily extended to study other disciplines and research areas without having to rely on ad-hoc science classification schemes.