M.S. Applied Data Science - Capstone Chronicles 2025

8

phone query but not the reverse, as users searching for a case are assumed to already own a phone. Furthermore, the authors assumed the notion of relatedness to be subjective and product dependent. To solve this orthogonal problem, a novel graph neural network (GNN), DAMEON, was proposed, wherein the problem is formulated as a node recommendation task on a directed product graph. The results, validated through offline experiments, showed DAMEON significantly outperformed contemporary model baselines. Notably, Rossi et al. (2022) confirmed the importance of directionality in their paper, “Edge Directionality Improves Learning on Heterophilic Graphs,” demonstrating that edge directionality enhances performance in certain cases. The authors suggested that due to early variants of GNNs, including directionality information was not widely adopted; however, directed GNNs (Dir-GNNs) have since been shown to outperform more complex models, specifically on connected nodes that tend to have different labels. From Clicks to Context: A Heterogeneous Graph Framework for Diagnosing Consumer Shopping Goals and Personalizing Retail Strategy Even with these advancements, translating graph model results into practical managerial insights can be challenging in retail environments. To address this disconnect, Yan and Xie (2025) proposed a heterogeneous graph framework that diagnoses the underlying drivers of network behavior by quantifying the influence of each contextual factor. This approach improves recommendation accuracy, return on investment, and downstream managerial buy-in, establishing graph analytics as a key strategic decision-making tool. Ultimately, while modern graph-based methods provide powerful tools to represent the interconnected landscape of brands, products, and consumer behaviors, translating these insights into clear, strategic decisions remains an ongoing challenge. Future research should prioritize

216