Explainable recommendation aims to develop models that not only provide high-quality recommendations but also generate intuitive explanations. These explanations can be post-hoc or directly come from an explainable model. The goal is to help users or system designers understand why certain items are recommended, thereby improving the transparency, persuasiveness, effectiveness, trustworthiness, and satisfaction of recommendation systems. It also aids system designers in debugging and refining algorithms. This survey provides a comprehensive review of explainable recommendation research. It categorizes recommendation problems into the 5W framework (what, when, who, where, why), highlighting the role of explainable recommendation in addressing the "why" aspect. The survey covers three perspectives: a chronological timeline of research, a two-dimensional taxonomy of explanations (information source and algorithmic mechanism), and the application of explainable recommendation in various tasks. Explainable recommendation models include model-intrinsic and model-agnostic approaches. Model-intrinsic models are interpretable, while model-agnostic models use post-hoc explanations. The survey discusses various methods, such as factorization models, topic modeling, graph-based models, and deep learning for explainable recommendation. Evaluation of explainable recommendation includes user studies, online and offline evaluations, and qualitative case studies. The survey also explores the application of explainable recommendation in different domains, such as e-commerce, point-of-interest, social, and multimedia recommendation systems. The survey identifies open directions and new perspectives in explainable recommendation, including methods and new applications, evaluation and user behavior analysis, explanation for broader impacts, and cognitive science foundations. It concludes by emphasizing the importance of explainable recommendation in improving the transparency and trustworthiness of recommendation systems.Explainable recommendation aims to develop models that not only provide high-quality recommendations but also generate intuitive explanations. These explanations can be post-hoc or directly come from an explainable model. The goal is to help users or system designers understand why certain items are recommended, thereby improving the transparency, persuasiveness, effectiveness, trustworthiness, and satisfaction of recommendation systems. It also aids system designers in debugging and refining algorithms. This survey provides a comprehensive review of explainable recommendation research. It categorizes recommendation problems into the 5W framework (what, when, who, where, why), highlighting the role of explainable recommendation in addressing the "why" aspect. The survey covers three perspectives: a chronological timeline of research, a two-dimensional taxonomy of explanations (information source and algorithmic mechanism), and the application of explainable recommendation in various tasks. Explainable recommendation models include model-intrinsic and model-agnostic approaches. Model-intrinsic models are interpretable, while model-agnostic models use post-hoc explanations. The survey discusses various methods, such as factorization models, topic modeling, graph-based models, and deep learning for explainable recommendation. Evaluation of explainable recommendation includes user studies, online and offline evaluations, and qualitative case studies. The survey also explores the application of explainable recommendation in different domains, such as e-commerce, point-of-interest, social, and multimedia recommendation systems. The survey identifies open directions and new perspectives in explainable recommendation, including methods and new applications, evaluation and user behavior analysis, explanation for broader impacts, and cognitive science foundations. It concludes by emphasizing the importance of explainable recommendation in improving the transparency and trustworthiness of recommendation systems.