M.S. Applied Data Science - Capstone Chronicles 2025

7

segmentation ability and providing actionable, data-driven insights that businesses can use to tailor marketing campaigns and inform future growth strategies in the digital market. Our study expanded on this by employing other unsupervised clustering techniques on physical cross shopping networks rather than social media networks. Graph-Based Approach for Industrial Market Segmentation Using Financial Transaction Network Structure and Firm’s Attribute Clustering techniques have been employed well in digital environments; however, there are alternative methodologies that can better articulate behavior based on scenario. Bose and Kim (2025) proposed a novel methodology to address the complexities of business-to-business market segmentation, specifically in financial transaction data. The study highlighted how traditional clustering methods fail to handle the massive, intricate, and often overlapping nature of industrial segments. To overcome this, they modeled the market as a financial transaction network, where firms are nodes and transactions are weighted edges, incorporating attributes like SIC codes alongside network topology. By integrating a graph-based, generative probabilistic model, the study successfully identified clusters that reflect the complex and overlapping relationships between firms. The results, validated against a real-world dataset, demonstrate that their proposed network-based community-detection method significantly outperforms conventional clustering and community-detection techniques, offering a more granular and accurate way for businesses to allocate resources and tailor marketing strategies. Recommending Related Products Using Graph Neural Networks in Directed Graphs In a related segmentation problem, Virinchi et al. (2023) discussed related product recommendation in the consumer space. This problem assumed asymmetry in that for a given product, there is a directional component. For example, it is suitable to recommend a case for a

215

Made with FlippingBook flipbook maker