随着大数据时代的到来，基于多源多模态数据的知识挖掘对企业日益重要。现阶段以深度学习为代表的机器学习技术为企业基于多源多模态数据的知识挖掘提供了可能和手段，已在各行各业得到了广泛应用。但其对学习过程和结果难以做出解释，严重制约了企业对深度学习技术的采纳和应用。因此，可解释的深度学习是未来的发展趋势，以深度学习技术为主流的知识挖掘同样呼唤可解释性。本项目旨在运用知识图谱技术与动态自适应的深度学习技术，提出一套在知识挖掘过程中基于全生命周期的动态可解释性深度学习方法，提高模型的准确率和效率，增强决策的科学性和可解释性。主要目标有三个：1.提出一套在知识挖掘过程中基于全生命周期的动态可解释性深度学习算法；2.建立一套工具，通过知识图谱技术将领域知识和元数据知识进行表达和融合，实现数据特征、模型指标和结果的可视化展示；3.在金融、医疗、汽车三个行业对所提出的算法和工具进行应用，评估其效果和可行性。

With the arrival of the era of big data, knowledge mining based on multi-source and multi-modal data is becoming increasingly important for enterprises. At present, machine learning technology represented by deep learning provides the possibility and means for enterprise knowledge mining based on multi-source and multi-modal data, and has been widely used in various industries. However, it is difficult to explain the learning process and results, which seriously restricts the adoption and application of deep learning technology in enterprises. Therefore, interpretable deep learning is the trend of development in the future, and knowledge mining with deep learning technology as the mainstream also calls for interpretability. This project aims to use knowledge mapping technology and dynamic adaptive deep learning technology to propose a set of dynamic interpretable deep learning methods based on the whole life cycle in the process of knowledge mining, to improve the accuracy and efficiency of the model, and to enhance the interpretability and scientificity of decision-making. There are three main objectives: 1. To propose a dynamic interpretable deep learning algorithm based on the whole life cycle in the process of knowledge mining. 2. Establish a set of tools to express and integrate domain knowledge and metadata knowledge through knowledge mapping and other technologies, and realize the visualization of data characteristics, model indicators and results. 3. Apply the proposed algorithms and tools in the financial, medical and automotive industries to evaluate their effectiveness and feasibility.