关键词: 人工智能, 创新链, 产学研协同网络, 指数随机图模型, 工业机器人

Abstract: Analyzing the structural bottlenecks and formation mechanisms of the industry-university-research (IUR) collaboration network in artificial intelligence (AI) from an innovation chain perspective is of great significance for enhancing the efficiency of the industrial innovation system and cultivating new quality productive forces. Using the industrial robot industry as a typical sample of AI applications in manufacturing, this paper constructs IUR collaboration networks across three stages—science, technology, and commercialization respectively, based on multi-source data from 2008 to 2024. It employs social network analysis (SNA) and exponential random graph models (ERGM) to analyze their structural characteristics and formation mechanisms. The findings reveal that the industrial innovation chain presents an unbalanced pattern characterized by a “wide entry but narrow passage”. Specifically, the scientific collaboration network exhibits highly cohesive “small-world” characteristics, the technological network displays a fragmented structure, and the commercialization network manifests as a fragile architecture reliant on a few platform-based hub nodes, with weak connectivity between stages. Innovation actors follow a role evolution path along the innovation chain, transitioning from “academia/research-led” to “enterprise-led” and further to “enterprise-core”. The network formation mechanisms exhibit significant stage heterogeneity: degree-based attachment drives the scientific network from diffusion towards clustering, reinforces the decentralization trend in the technological network, and reduces the decentralization tendency in the commercialization network. Actor type matching shows a homophilous synergy state in the scientific stage but shifts to heterophilous complementarity in the technology and commercialization stages. Triadic closure and innovation capability, as core forces running through the entire process, possess structural cohesion and innovation appeal, continuously promoting the formation of collaborative relationships. This paper reveals the asymmetric evolution patterns and connection bottlenecks of the three-stage AI IUR collaboration network from the dual dimensions of structure and mechanism, providing empirical evidence for improving the operational efficiency of the innovation chain and promoting differentiated stage-specific governance.

Key words: Artificial Intelligence, Innovation Chain, Industry-University-Research Collaboration Network, Exponential Random Graph Model, Industrial Robot