Governing innovation networks under uncertainty: a strategic framework for resilient knowledge flows

Amid innovation uncertainty, maintaining knowledge-flow reachability between firms is critical to network governance. While existing research has largely focused on assessing network vulnerability or predicting collaborative relationships, few studies have explored how to design resilience-enhancing collaborative connections proactively under resource constraints. Based on joint patent data from the new energy vehicle industry, this study proposes the GNN-Assist resilience intervention framework. Unlike traditional link prediction, which focuses on the probability of relationship formation, GNN-Assist treats candidate collaborative edges as governable intervention options, evaluating their marginal contribution to structural knowledge accessibility under scenarios of random failure and core node removal. The framework integrates graph learning prediction, cross-community bridging, de-core dependencies, and budget constraints to identify potential connections with higher resilience values. Results show that GNN-Assist achieves the most stable comprehensive resilience gains, with an average ΔRavg of 0.0204-an improvement of approximately 19% over leading link prediction benchmarks and approximately 42% over the best structural heuristic methods. These advantages are particularly evident during core node removal and the early stages of shocks. This study shifts network resilience research from ex-post evaluation to ex-ante intervention design, offering a structural tool for prioritizing collaborations.