Modularizing artefact knowledge promotes technological impact

Technological evolution depends not only on the invention of new artefacts but also on how their knowledge is structured, represented, and propagated. In this study, we examine how the modularity of artefact knowledge influences technological impact. We utilize a dataset of 33,803 patents from the United States Patent & Trademark Office (USPTO) and their knowledge graphs constructed using the facts extracted from patent descriptions. Using a regression analysis controlling for several structural properties of the knowledge graphs, we establish a significant positive relationship between modularity of the graph structures—measured using the Louvain method and the technological impact, as quantified by normalized forward citations. To further examine this relationship, we develop a predictive framework integrating Graph Neural Networks (GNNs) and regression models to estimate normalized citation scores from patent knowledge graphs. We then apply this framework to conduct a counterfactual analysis, wherein, we tune the modularity of knowledge graphs and assess the enhancement in expected citations. The analysis reveals that patents with less or no citations could benefit the most from modularization of their knowledge, as a citation gain could help initiate their knowledge propagation. We also discuss with a few examples as to how re-representation of artefact knowledge is necessary in addition to re-designing artefacts for modularity.