Recommender systems (RSs), which provide recommendations tailored to user preferences, are valuable in managing overloaded information. Traditional recommendation systems usually function as black-box models and lack explanation; as a result, user trust and system transparency are adversely affected. Explainable RSs (XRSs) aim to overcome this issue by providing interpretable justifications for recommendations. Previous XRS studies suffer from limited integration of user reviews with knowledge graphs (KGs), resulting in incomplete user preference modeling and lack of interpretability. Although improvements in XRSs have been achieved worldwide, studies on Arabic RSs a lack advanced tools and explanation methods, such as KGs, because of resource limitations, the challenges posed by the Arabic language, and its different dialects. This study introduces Shareh, an explainable KG-based Arabic recommender that utilizes meta-path-guided reasoning and graph attention networks to fuse user reviews with a heterogeneous Arabic KG. Experimental results on the Books Reviews Arabic Dataset show that Shareh improves mean absolute error (MAE), and root mean squared error (RMSE) by approximately 30% compared to baseline models. The system, which has a model fidelity score of 99.76%, effectively backs nearly all recommendations with reasonable explanations. Such high fidelity indicates that the produced explanations accurately reflect the fundamental concepts of the model, thereby improving the system’s interpretability and reliability and increasing user trust.
Katheeth et al. (2026) studied this question.