<p>In large-scale complex knowledge graphs, existing approaches often overlook the hierarchical structure of data, limiting their effectiveness in applications such as information and recommender systems. To address this, we propose CKRHE, a hierarchical embedding technique that integrates entity types to capture the inherent hierarchy in knowledge graphs. Our method builds on translation-based embedding models and utilizes continuous bag-of-words and CNN architectures to improve semantic representation learning. A joint loss function is introduced to balance both descriptive and hierarchical losses, with adaptive equilibrium coefficients to optimize performance. Experimental results on benchmark and custom datasets demonstrate that CKRHE significantly outperforms existing methods in link prediction tasks, achieving 4.1% and 1.4% improvements in Hits@10 and MRR metrics, respectively. Our findings highlight the critical role of hierarchy-aware embedding in enhancing reasoning accuracy for large-scale knowledge graphs.</p>