Enhancing Consumer Store Choice and Cross-Purchase Modelling through Graph Machine Learning

Abstract

This thesis advances consumer behavior research by integrating computational techniques to model complex retail decision-making. It bridges marketing theory and practice, focusing on basket selection and store choice, while enhancing econometric models with machine learning.

Motivated by challenges in capturing product interdependencies and spatial influences, traditional models often overlook nuanced patterns. This work employs Graph Neural Networks (GNNs) and Long Short-Term Memory (LSTM) to address these gaps.

Study 1: Basket Selection

Uses GNNs for graph embeddings of product relationships, integrated into a conditional logit model to predict basket composition more accurately.

Study 2: Store Choice

Combines GNNs with LSTM attention to derive location features from purchase sequences, improving predictions of store preferences.

These innovations offer methodological advancements and practical tools for retailers to boost engagement and performance. The research enriches marketing science, paving the way for future explorations in data-driven consumer analysis.