Modularizing Patent Knowledge for Enhanced Technological Impact

Modularity is a core design principle that enables technological artefacts to remain adaptable and evolvable, while creating value, maintaining technical robustness, and protecting intellectual property. Despite the perceived advantages, modularity has mostly been theoretically argued without sufficient empirical support—primarily due to the lack of large-scale datasets of structured representations of the knowledge of technological artefacts. In this study, we leverage a recently populated dataset of over 33,800 patent knowledge graphs built by recurrently extracting facts (head entity-relationship-tail entity) from patent sentences. Considering these knowledge graphs as explicit representations of patent knowledge, we investigate the influence of modularity upon the technological impact of patents, controlling for structural and semantic variables of graphs. We find a consistent positive influence of modularity on technological impact—quantified by short-term (5 years) and long-term (10 years) forward citation scores. Empirically substantiating the influence of modularity argued in design theories, we develop a predictive framework combining Graph Neural Networks (GNNs) and regression models to estimate citation scores from patent knowledge graphs. Using this framework, we showcase how modifications to patent knowledge—either through re-design or re-representation—can enhance the citation scores increasingly over 5- to 10-year periods—particularly for under- or un-cited patents.