Uncovering the structural dynamics of research collaboration in computer science: an information entropy perspective

Evaluating the structural evolution of research collaboration is critical for understanding scientific progress and optimizing research policy. This study adopts an information entropy perspective to quantify the static features and dynamic trajectories of collaboration networks. Using DBLP publication data from 2005 to 2024, we constructed annual weighted co-authorship networks to investigate the collaboration ecosystem within the Computer Science domain and analyzed them using three normalized metrics: degree entropy, clustering coefficient entropy, and average path length entropy. Findings reveal that collaboration networks maintain a highly ordered core-periphery structure, evidenced by consistently low degree entropy. A significant linear growth in clustering coefficient entropy indicates structural polarization, where the network evolves into a hybrid of dense communities and sparse peripheral nodes. Furthermore, path length entropy exhibits an inverted U-shaped pattern, due to structural homogenization driven by ‘super-connectors’. These patterns suggest the network has matured into an ecosystem that balances innovation modes through structural differentiation. High frequency ‘super-pairs’ facilitate exploitative innovation, while ‘super-connectors’ enable exploratory breakthroughs. For research evaluators, these entropy-based metrics provide a robust quantitative framework to monitor network maturity and innovation capacity. We conclude that structural polarization, rather than simple expansion, is a key mechanism for maintaining high-efficiency scientific innovation.