Building a Domain-Specific Knowledge Graph for Business Networking Analysis

The emergence of graph learning supports the study of phenomenon that are best represented by knowledge graphs, using deep learning techniques. However, deep learning is mostly built around supervised learning which requires datasets for training and testing. This paper presents BusinessNet, a domain-specific knowledge graph to examine the business networking phenomenon and offers a novel dataset for the graph learning community. BusinessNet is built from unstructured news wire text data from which entities such as organizations, people, and places, as well as the relationships among them are extracted to build the graph. The resulting graph is a heterogeneous network of organizations, people, and places as nodes and that will include different types relations. The analysis of the resulting dataset shows interesting characteristics for graph learning models evaluations.