RippleNet is an end-to-end framework that incorporates knowledge graphs into recommender systems to propagate user preferences. It addresses the sparsity and cold start problems in collaborative filtering by using knowledge graph (KG) information. RippleNet simulates the propagation of user preferences through the KG, extending users' potential interests iteratively along KG links. This process generates multiple "ripples" that superpose to form a user's preference distribution over candidate items, which can be used to predict the final clicking probability. Through extensive experiments on real-world datasets, RippleNet achieves significant improvements in various recommendation scenarios, including movies, books, and news, outperforming state-of-the-art baselines. RippleNet combines the advantages of embedding-based and path-based methods, allowing it to automatically discover potential paths from a user's history to a candidate item. It also provides a new perspective on explainability by tracking paths from a user's history to an item with high relevance probability. The framework uses a Bayesian approach to unify preference propagation with KGE regularization for click-through rate prediction. RippleNet's performance is evaluated on three real-world scenarios, showing its effectiveness in top-K recommendation and CTR prediction. The results demonstrate that RippleNet outperforms other methods in terms of AUC and other metrics, with significant improvements in movie, book, and news recommendation. The framework is also sensitive to parameters such as embedding dimension and KGE regularization strength, with optimal performance achieved at specific values. Future work includes further investigation into entity-relation interactions and designing non-uniform samplers during preference propagation.RippleNet is an end-to-end framework that incorporates knowledge graphs into recommender systems to propagate user preferences. It addresses the sparsity and cold start problems in collaborative filtering by using knowledge graph (KG) information. RippleNet simulates the propagation of user preferences through the KG, extending users' potential interests iteratively along KG links. This process generates multiple "ripples" that superpose to form a user's preference distribution over candidate items, which can be used to predict the final clicking probability. Through extensive experiments on real-world datasets, RippleNet achieves significant improvements in various recommendation scenarios, including movies, books, and news, outperforming state-of-the-art baselines. RippleNet combines the advantages of embedding-based and path-based methods, allowing it to automatically discover potential paths from a user's history to a candidate item. It also provides a new perspective on explainability by tracking paths from a user's history to an item with high relevance probability. The framework uses a Bayesian approach to unify preference propagation with KGE regularization for click-through rate prediction. RippleNet's performance is evaluated on three real-world scenarios, showing its effectiveness in top-K recommendation and CTR prediction. The results demonstrate that RippleNet outperforms other methods in terms of AUC and other metrics, with significant improvements in movie, book, and news recommendation. The framework is also sensitive to parameters such as embedding dimension and KGE regularization strength, with optimal performance achieved at specific values. Future work includes further investigation into entity-relation interactions and designing non-uniform samplers during preference propagation.