This review summarizes the current state of drug-drug interaction (DDI) prediction using deep learning and knowledge graph (KG) techniques. DDIs can lead to unpredictable pharmacological effects and adverse events, making accurate prediction crucial. Traditional methods are time-consuming and expensive, so computational approaches are needed. Deep learning and KG techniques are effective for feature extraction and have been widely used in DDI prediction. The review categorizes existing methods into three classes: deep learning-based, KG-based, and hybrid methods combining both. It discusses the challenges in DDI prediction, including asymmetric and high-order interactions. DDI prediction involves collecting drug-related data from public databases such as DrugBank, KEGG, PubChem, and DRKG. These databases provide information on drug structures, targets, enzymes, and adverse effects. Benchmark datasets are used to evaluate prediction methods, and various models are compared on these datasets. The review highlights the use of machine learning and deep learning techniques, including deep neural networks (DNN), graph neural networks (GNN), autoencoders (AE), and KG-based methods. These models are evaluated for their ability to predict DDIs, with a focus on accuracy, interpretability, and performance on benchmark datasets. Deep learning models, such as DNN and GNN, are effective in capturing complex patterns in high-dimensional data. GNNs, in particular, are useful for modeling drug interactions as graphs, where nodes represent drugs and edges represent interactions. Autoencoders are used for dimensionality reduction and feature extraction. KG-based methods leverage structured knowledge to enhance prediction accuracy. Hybrid methods combine the strengths of deep learning and KG to improve performance. The review also discusses the challenges in DDI prediction, including the need for more comprehensive data, handling imbalanced datasets, and improving model interpretability. Future research should focus on integrating multi-modal data, enhancing model robustness, and improving the interpretability of deep learning models to better understand the mechanisms behind DDIs. Overall, the integration of deep learning and KG techniques offers promising approaches for improving the accuracy and efficiency of DDI prediction.This review summarizes the current state of drug-drug interaction (DDI) prediction using deep learning and knowledge graph (KG) techniques. DDIs can lead to unpredictable pharmacological effects and adverse events, making accurate prediction crucial. Traditional methods are time-consuming and expensive, so computational approaches are needed. Deep learning and KG techniques are effective for feature extraction and have been widely used in DDI prediction. The review categorizes existing methods into three classes: deep learning-based, KG-based, and hybrid methods combining both. It discusses the challenges in DDI prediction, including asymmetric and high-order interactions. DDI prediction involves collecting drug-related data from public databases such as DrugBank, KEGG, PubChem, and DRKG. These databases provide information on drug structures, targets, enzymes, and adverse effects. Benchmark datasets are used to evaluate prediction methods, and various models are compared on these datasets. The review highlights the use of machine learning and deep learning techniques, including deep neural networks (DNN), graph neural networks (GNN), autoencoders (AE), and KG-based methods. These models are evaluated for their ability to predict DDIs, with a focus on accuracy, interpretability, and performance on benchmark datasets. Deep learning models, such as DNN and GNN, are effective in capturing complex patterns in high-dimensional data. GNNs, in particular, are useful for modeling drug interactions as graphs, where nodes represent drugs and edges represent interactions. Autoencoders are used for dimensionality reduction and feature extraction. KG-based methods leverage structured knowledge to enhance prediction accuracy. Hybrid methods combine the strengths of deep learning and KG to improve performance. The review also discusses the challenges in DDI prediction, including the need for more comprehensive data, handling imbalanced datasets, and improving model interpretability. Future research should focus on integrating multi-modal data, enhancing model robustness, and improving the interpretability of deep learning models to better understand the mechanisms behind DDIs. Overall, the integration of deep learning and KG techniques offers promising approaches for improving the accuracy and efficiency of DDI prediction.