Unraveling technology diffusion through dynamic network and multi-dimensional mechanism analysis: evidence from natural language processing

Introduction. Understanding factors influencing technology diffusion is vital for optimizing technological environments and fostering innovation. Existing studies often overlook temporal dependence and lack multidimensional mechanism analysis. This study addresses these gaps by introducing a dynamic network perspective to analyze technology diffusion. Method. We developed a framework that integrates topic extraction with dynamic relationship modeling. Using patent data, BERTopic was applied to identify technological topics and construct cross–time-slice diffusion networks. Social network analysis captured evolutionary patterns, while the temporal exponential random graph model (TERGM) jointly examined endogenous network structures, actor–relation effects, and exogenous factors. Analysis. The natural language processing field was selected as a case study. Diffusion dynamics and mechanism factors were investigated through quantitative modeling of temporal networks. Results. The network has become more cohesive yet decentralized. Core nodes remain but their bridging role weakens. Reciprocity strongly promotes diffusion. Topic influence, novelty, and knowledge quality positively drive relationship formation, while knowledge breadth and depth affect only the sender effect. Conclusions. This study integrates dynamic networks with multidimensional mechanism analysis, bridging gaps in temporal evolution and mechanism exploration, and providing a reusable framework and empirical reference for technology diffusion research.