Answers are grounded in this paper and cite their sources.
Computational study demonstrates accurate entity classification and link prediction across innovation ecosystems, indicating the power of heterogeneous graph modeling.
The innovation and entrepreneurship ecosystem consists of multiple stakeholders, including governments, universities, and enterprises, and involves complex interactions among technological cooperation, investment relations, and policy support. To improve the representation and prediction capability of multi-entity and multi-relation ecosystem models, this paper introduces a heterogeneous graph neural network framework covering graph construction, feature representation, and relation-embedding modeling. A heterogeneous graph is first constructed with enterprises, universities, and governments as node types and technological cooperation, investment, and policy support as edge types. Structural and behavioral features of different nodes are then extracted and uniformly encoded into multimodal vectors. A heterogeneous graph attention network is subsequently introduced to conduct weighted modeling of node types and relationship semantics. Finally, low-dimensional node representations are generated through graph embedding for ecosystem subgroup partitioning, core node identification, and potential relationship prediction. Experimental results show classification accuracies of 0.92, 0.91, and 0.95 for enterprise, university, and government nodes, respectively. The AUC values for link prediction in technological cooperation, investment, and policy support relations all exceed 0.92. Node centrality correlations and semantic separation results further verify the effectiveness of multimodal embedding in preserving network structure and distinguishing heterogeneous entities.
No takes yet. Share an insight, caveat, or question.
W. W. Yuan (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: