Heterogeneous business network based interpretable competitive firm identification: a graph neural network method

Abstract

Competitor identification at the firm-level is a crucial aspect of business strategy. In general, we can identify competitive firms based on market commonality and resource similarity. However, previous studies predominantly focus on market commonality, seldom consider the resource similarity. To bridge this gap, we attempt to use three types of business relationships, namely, customer relationships, supplier relationships and alliance relationships to identify competitors from the perspective of resource similarity. First, we consider the connectivity and heterogeneity of business relationships and build a heterogeneous business network (HBN) to transform hybrid and complex business relationships into a heterogeneous information network. Then, to identify competitors from the HBN and improve the interpretation of prediction results, we further develop a heterogeneous graph neural network based competitive firm identification (HGNN-CFI) method. Finally, extensive experiments on 3371 firms reveal that HGNN-CFI can not only identify competitors but also offer interpretability for the prediction results. This paper provides a novel perspective and method to identify competitive firms.