<p>Analyzing complex bibliographic networks is essential for understanding research trends and collaborations. Heterogeneous graph neural networks (GNNs) are adept at processing diverse information but are resource-intensive. Conversely, homogeneous GNNs are more efficient but struggle to accurately represent different types of information, making deep thematic representation difficult. To address this, we propose heterogeneous semantic projection optimization (HSPO), a novel approach to enhance thematic depth within homogeneous GNNs. Our method projects rich semantic information from heterogeneous bibliographic networks onto homogeneous networks, enabling more detailed and efficient analysis. We included an evaluation step to verify the method’s effectiveness in predicting future research areas based on the analyzed bibliographic network. Experimental results demonstrated that our method increased the F1-score for single-category recommendations by an average of 10.33% compared to existing GNNs. Additionally, our method significantly improved performance by reducing training time by 59.10% compared to the heterogeneous graph transformer (HGT). These findings confirm the effectiveness of the HSPO approach in optimizing thematic depth while maintaining computational efficiency.</p>