Abstract
To enhance universities' risk identification and response capabilities in complex network environments, this study proposes a university network security management model integrating transfer learning and multi-agent collaborative governance mechanisms. It aims to construct an implementable intelligent governance system. Methodologically, this study develops an attention-enhanced transfer residual network that incorporates a channel attention mechanism to strengthen feature selection. The network achieves deep cross-domain alignment through integration with the maximum mean discrepancy method, significantly enhancing recognition accuracy under conditions of limited target domain data availability. Subsequently, a school-family-community collaborative response mechanism based on multi-agent game theory is constructed. The mechanism drives the generation of personalized intervention strategies through a risk scoring function to address response lags and responsibility ambiguity in traditional governance. Results show that the proposed model achieves an accuracy of 95.7%, a recall of 94.8%, a F1 score of 0.951, an Area Under the Curve of 0.973, and detection delay controlled within 29 ms. The integrated model proposed in this study demonstrates strong identification capability.