Files
Abstract
Knowledge recombination theory has been widely accepted and applied across disciplines, including sociology, management, and data science. However, its underlying promise of a generalized explanation has been challenged theoretically and lacks consistent empirical support. This project therefore explores the mechanisms through which innovation emerges, and how these mechanisms unfold in unique social contexts. Drawing on five large-scale datasets from key knowledge production systems—science, technology, business strategy, art, and online knowledge platforms—I use advanced statistical methods and computational tools to discuss the generality and specificity of the recombination theory across these domains.
The three chapters focus on different dimensions of combinatorial innovation. The first chapter closely examines invention in the context of entrepreneurship, drawing on theories of organizational emergence and strategic positioning literature. We find that startups perform best when they combine several mature technological applications within the existing economic system, rather than creating entirely new application scenarios. This study identifies risk distribution as a key predictor of recombination outcomes. Startups that mobilize “higher-order invention” —invention across subsystems rather than within them—receive disproportionate market rewards. Meanwhile, preferences for risk distribution may differ across other systems.
The second chapter uses patent data to examine combinatorial trajectories in technology. I first constructed a technological map using Poincaré embedding, a representational learning algorithm designed to capture hierarchical structures. This map effectively reflects the architecture of the knowledge tree. Based on knowledge trajectories mapped onto this space, I find that deep search—combining closely related knowledge—gains rapid attention within the community, while broad search—linking distant elements—accumulates more attention over time by attracting a wider audience. This study highlights the availability of knowledge components as a key dimension of combinatorial innovation: the more intuitive the combination, the lower the risk, but also the lower the potential for disruptive innovation.
Building on findings from the first two chapters, the final chapter turns to the classic topic of team design. In this study, we distinguish between two dimensions of diversity in teams: background diversity—the differences in team members’ knowledge backgrounds, which influences the availability and readiness of knowledge components; and perspective diversity—the differences in how team members interpret the same issue, which affects the distribution of cognitive risk. We find that teams with low background diversity and high perspective diversity consistently perform better across all five domains. This study illustrates how a generalizable social law in innovation can be identified while still accounting for contextual particularities.
This research advances theories in the sociology of knowledge, group processes, and organizational studies. For sociology of knowledge, particularly theories related to science and technology, it highlights the boundary conditions of recombination theory. For group process theories, it introduces a new concept and analytical framework for assessing knowledge complementarity. For organizational science, it proposes several future directions for exploring new dimensions of ambidexterity.