The rapid expansion of the silicon carbide (SiC) semiconductor industry has created vast and heterogeneous data streams from academic publications, patents, technical standards, and market reports. Traditional methods struggle to integrate these multi-source datasets, limiting their ability to reveal technological trends and innovation pathways. To address this challenge, we propose an Information-Enhanced Heterogeneous Graph Convolutional Network (ie-HGCN) framework for multi-source data fusion in the SiC semiconductor electronic information industry. The method combines principal component analysis and random forest algorithms for feature extraction with a DeepGCNs-Att architecture augmented by self-attention mechanisms, enabling the effective modeling of complex relationships among heterogeneous data entities. Experimental results show that ie-HGCN outperforms conventional deep learning and graph-based baselines, achieving 88.63% accuracy, 90.65% precision, 87.54% recall, and an F1-score of 85.68%. A case study on patents and publications demonstrates the framework’s ability to identify emerging hotspots, such as advanced power module packaging, and to uncover novel industry-academia collaborations. These findings highlight the practical value of ie-HGCN as a robust tool for strategic R&D planning and technology forecasting in the SiC semiconductor domain.
YiDong Zhu (Sun,) studied this question.