研发项目是涉及顾客需求、产品功能和部件、团队等多知识领域的复杂系统，与大数据技术相关的知识图谱方法可以更加客观全面地展示、分析不同领域间的关联，为此，本文提出新产品开发(NPD)知识图谱，并将其与依赖结构矩阵(DSM)等方法相结合，以识别研发项目中多领域间的相互依赖关系。首先，本文建立依据NPD知识图谱测度顾客需求优先序的模型，并采用DSM和质量功能展开(QFD)方法，建立由“需求-功能”QFD关联推导功能间依赖关系强度的模型。然后，采用“功能-产品”多领域矩阵(MDM)推导部件间的依赖关系强度。最后，对DSM进行聚类，为提高聚类算法的稳定性，采用改进的信息熵，建立了改进的基于信息熵的两阶段DSM聚类模型，算例分析表明，该方法可明显降低类间的协调复杂性并提高算法的稳定性。

   The new product development (NPD) project is a complex system involving customer needs, product functions and components, organization and other different knowledge fields. These domains or fields interact and influence each other. NPD is undergoing a transformation from "technology-centered" to "customer-centered". Hence, how to integrate customer knowledge into the product development and innovation process has become more and more important. The knowledge graph method related to big data can more objectively and comprehensively analyze the dependency relationship between different domains than traditional methods and dependency structure matrix (DSM) is a useful tool to analyze the relationship between multiple knowledge domains. The knowledge graph method uses the visual method to identify and construct the relationship between entities and attributes, which can more truly show the relationship between entities in the real world.
   Accurate identification of customer needs is basic challenge for developing new products. Knowledge graph related to big data technology can effectively solve this problem. Based on information such as customers′ online behavior, knowledge graph technology can accurately capture customers′ needs and their priorities. At the same time, the knowledge graph can more comprehensively and visually reflect the complex relationships among multiple knowledge domains, such as customer needs, products and organizations involved in the research and development project, and quantify the internal and external relationships of each domain, thus greatly improve the accuracy of data and NPD project management efficiency. However, previous studies on knowledge graph mainly focus on the technical aspects such as how to construct the knowledge graph, but ignore the application of knowledge graph in NPD projects. 
    Thus, this paper proposes the NPD knowledge graph and combines it with dependency structure matrix (DSM) to identify the dependence strength among different domains in NPD projects. First, according to the relationship between customer needs and customers in the NPD knowledge graph, this paper builds models to measure the priority of customer needs. We measure the weight of customer needs according to the number of links between customer needs and customer nodes. Next, combing the DSM and Quality Function Deployment (QFD) method, this paper builds models to derive the dependency strength between functions through the "demand-function" QFD. The dependency between components includes direct and indirect dependency relationship. Then, the function-product DMM (Domain Mapping Matrix) is built to further analyze the dependency between elements in the function and product domains. And the "function-product" MDM (multi-domain matrix) is established through integrating the functional DSM, the product DSM, and function-product DMM. Using the "function-product" MDM, this paper builds models to derive the dependency strength between components through known DSM and MDM. Finally, to improve the algorithm′s stability and reduce the management complexity through the clustering method, this paper presents a two-stage clustering method based on minimizing the Weighted Average Entropy (WAE) which includes External Cluster Average Entropy (ECAE) and Internal Cluster Average Entropy (ICAE). The information entropy represents the order degree of information coordination between elements of internal cluster or external cluster and it is related with the possibility of information exchange.
    An industrial example is used to illustrate the proposed method and model. The results reinforce several managerial practices but also yield new insights, such as how to measure the priority of customer needs, how to derive the dependency strength between components using the "function-product" MDM, and how to clustering the organization based on minimizing the WAE. The example results show that the proposed method can reduce the coordination complexity between clusters and improve the algorithm′s stability.