M.S. Applied Data Science - Capstone Chronicles 2025

6

A Product-Network Analysis for Extending the Market Basket Analysis While early probabilistic models captured latent structures statistically, subsequent research began to model these relationships explicitly as networks, revealing a more interconnected view of market competition. Kim et al. (2012) proposed a product-network analysis to emphasize the network-leveled relationship between products, including a copurchased product network. Interestingly, the authors found that bestsellers do not necessarily occupy central positions in networks, and products with higher centrality tend to be more effective in sales promotion and increase sales volumes. This research inspired us to develop a network model based on cross-shopping at physical locations across brands, rather than products. Although Kim et al. discovered the bestsellers do not necessarily occupy central positions, we hypothesized the presence of central hubs in our graph models. Data-Driven Strategies for Digital Native Market Segmentation Using Clustering More recently, Uddin et al. (2024) addressed the challenges faced by marketers in understanding the evolving behavioral and psychological patterns of “digital native” consumers, particularly teenagers, on social networking sites. This study expanded the network perspective by applying product-interaction theory to contemporary consumer habits. The study proposed a machine learning-based digital native market segmentation model that leveraged an open-access prototype dataset from social networking site platforms. By employing and evaluating various clustering techniques—including K-means, MiniBatch K-means, AGNES, and Fuzzy C means—the research successfully uncovered hidden interests and grouped consumers with similar tastes. The findings highlight the superior performance of the K-means algorithm in this specific context, achieving a 63.90% silhouette score for two clusters, demonstrating its strong

214