Prediction of Business Partners Using an N-Gram-Based Approach that Combines a Network Model and Linear Model of a Supply Chain

Supply chains are viewed as networks of both goods and services, and knowledge and information. Their knowledge will be potential resources for new business relationship with hidden partners; however, many companies find it difficult to develop new opportunities. A recommendation of a potential partner is helpful for regional revitalization. Research into supply chains has shifted from a linear model to a network model. A network model using graph theory can topologically explain a supply chain, whereas, in a linear model, there are some differences with the real world because of a lack of information. In this study, we propose a prediction model for a new business partnership with predictors extracted from network and linear approaches used in combination for the prediction of performance and interpretability. Our dataset consisted of a network of 327,012 transactions among 131,192 companies in Northeast Japan, which was retrieved from supplier-customer relationship data provided by Teikoku Databank, Ltd. Network centralities were extracted as topological features from the network of each company. Trigram relationships were also extracted from the network motif so that the logistic flow to a company from its supplier could be used to predict customers as business partners. The results showed that the performance of the proposed model was excellent, with a high contribution of probability extracted from the trigram relationships. From this perspective, we found that the information of logistics flows is a critical factor for predicting a potential partner, even in a network model.