Analyzing Enterprise Attribute-Dependent KPIs/KGIs by Bayesian Network-Leveraging LDA

Abstract

In product development, key performance indicators (KPIs) and key goal indicators (KGIs) have complex influences on each other. To understand the structure among them, Bayesian network analysis is one of effective methods. However, relationships among KPIs/KGIs often differ in attributes of enterprises, such as business type and annual sales. In this study, the authors incorporate topics obtained via latent Dirichlet allocation (LDA) into Bayesian network as nodes. With this “Bayesian network with topic nodes,” how KPIs affect the results of KGIs can be probabilistically inferenced and graphically observed according to attributes of enterprises. Furthermore, by configuring cultural or national differences as topic nodes, the proposed methods are expected to contribute overcoming barriers caused by these differences and accelerating improvement of product development in the global society.