Abstract:

Collaborative network is an important way for enterprises to adapt to the external uncertainty environment and conduct open innovation. Based on the joint patent application data in the SoC chips industry from 2003 to 2021, this paper analyzes, in detail, the topology and characteristics of the collaborative innovation network by using the social network analysis method, and empirically interprets the interaction mechanism and promotion paths of heterogeneity characteristics in various collaboration contexts on the innovation performance of enterprises by using the CART decision tree, K-means, and other machine learning methods. The results show that in the simple binary relationship networks, the innovation performance is mainly affected by the cooperation depth, while in the complex collaborative relationship networks, it is affected by both structural and non-structural factors. Betweenness centrality is the main factor. Cooperation depth and eigenvector centrality positively affect the innovation performance in those firms with weak resource bases. Degree centrality positively affects the innovation performance in those platform firms with concentrated resources. The effects of other features differ less in this two collaborative networks and do not have significant impacts on the innovation performance. The findings provide a path reference for relevant technology research and development enterprises to make partner selection and rationalize network resources to further improve innovation performance, and expand the analysis perspective for paradigm exploration of social science research in the era of digital economy.

Key words:

collaborative network, innovation performance, network structure, machine learning; decision rules

摘要：

合作网络是企业适应外部环境变革以进行开放式创新的重要方式。基于2003~2021年SoC芯片行业的联合专利申请数据，通过社会网络分析深入剖析了合作创新网络的拓扑结构及其异质性特征，并采用CART决策树、K-Means等机器学习方法实证分析了异质性网络情境中企业创新绩效的内在影响机制与提升路径。发现企业在简单二元关系网络中其创新绩效主要受到合作深度的影响，而在复杂合作关系网络中受到结构特征与非结构特征的共同作用。中介中心度是复杂合作环境中绩效的主要影响因素，合作深度与特征向量中心性在资源聚集程度低的企业群中正向影响创新绩效，度数中心度在资源密集的平台型企业群中正向影响创新绩效，其他特征在两种合作网络情境中差异较小，对创新绩效的影响并不明显。研究结论为相关技术研发企业进行合作伙伴选择、合理配置网络资源以提升创新绩效提供了网络嵌入路径参考，同时为数字经济时代社会科学研究的范式探索拓展了新的视角。

关键词:

合作网络, 创新绩效, 网络结构, 机器学习, 决策路径