Recommendation systems are extensively utilized across diverse domains with the aim of suggesting the most appropriate items to individual users. An effective recommendation engine provides valuable guidance for daily life. Therefore, improving the efficiency of recommender systems has become a prominent subject of interest in both industrial and academic circles. The majority of high-performing models in recommendation tasks utilize binary homogeneous graph convolutional methods and collaborative filtering as their foundational structures. These models are simple in structure, accurate and efficient in recommendation tasks. However, owing to the constraints of the binary homogeneous graph's structure, these models are unable to incorporate supplementary information during operation, thereby hindering the full utilization of diverse and efficient data. To fill the research gaps, this study introduces a novel recommendation system named REHG, which is grounded on directed heterogeneous graphs. Additionally, the study devises two experiments to assess its performance across various testing datasets. These experiments offer compelling evidence to support the superior performance of the proposed REHG model over common baseline models across different evaluation criteria. Furthermore, the study explores the feasibility of enhancing interoperability among various recommender system models.