<p>Unlike traditional networks, the hypernetworks have complex higher-order tuple relationships, i.e. hyperedges. However, most existing hypernetwork representation learning methods fail to adequately capture these complex higher-order tuple relationships. To address this issue, a vector knowledge transfer-driven representation learning method for heterogeneous hypernetworks abbreviated as VTRL is proposed. Firstly, a hyperedge-aware random walk algorithm with node importance preservation is proposed, which assigns greater random walk probabilities to more important nodes without decomposing the hyperedges. Secondly, based on hyperedge-aware random walk algorithm with node importance preservation, a hyperedge-aware topological structure model is proposed to learn pre-trained vectors. Finally, inspired by transfer learning, the pre-trained vectors are transfered to knowledge-enhanced similarity model to the final node representation vectors. The experiments on four real-world hypernetwork datasets demonstrate that as for link prediction tasks, our proposed method outperforms almost all baseline methods. As for hypernetwork reconstruction tasks, on the drug dataset, our proposed method outperforms all baseline methods when the reconstruction ratio is less than 0.8, meanwhile, on the GPS dataset, our proposed method shows competitive performance compared to the best baseline method HPHG.</p>