Data mining methodologies in interfirm transaction networks for business intelligence

Transaction network is a useful tool to represent transactional relationships between firms. As advances in technology facilitate connections between various entities, the study of complex networks has been increased in various fields, transaction networks being an example, to achieve diverse goals such as link prediction, clustering, and competitor identification. However, most existing research focuses on static undirected networks, and research on dynamic directed networks with attributes, such as transaction networks, is limited. In this dissertation, we present new machine learning methodologies that incorporate the characteristics of transaction networks. We first propose a link prediction procedure based on a new similarity score to predict new transactions using transaction networks. Our proposed score avoids the homophily assumption, common among existing methods, and exploits the hierarchical structure of transaction networks. The proposed method is tested on real-world transaction networks and yields larger area under the receiver operating characteristic curve (AUC) compared to existing methods. We then present a robust asymmetric non-negative matrix factorization method to cluster the nodes in directed networks. Recognizing that the errors do not follow Gaussian distributions in real-world networks, the proposed method assumes that the errors follow a Cauchy distribution, which resembles the Gaussian distribution but has heavier tails. Experiments using real-world as well as artificial networks show that the proposed method outperforms existing NMF methods and other representative work in clustering in various settings. In addition, we present a new method for competitor identification using financial transactions between firms. The proposed method quantifies the resource similarity and market commonality for pairs of firms in a transaction network. Same-level common demanders and suppliers and different-level ones are considered as well as edge information, number, and date of transactions. The proposed method is evaluated using financial transactions between firms from 2000 to 2015 in Korea. Finally, we present future research plans for the validity and further extensions of the works proposed in this dissertation.