Towards Preprocessing Guidelines for Neural Network Embedding of Customer Behavior in Digital Retail

Abstract

Shopping transactions in digital retailing platforms enable retailers to understand customers' needs for providing personalized experiences. Researchers started modeling transaction data through neural network embedding, which enables unsupervised learning of contextual similarities between attributes in shopping transactions. However, every study brings different approaches for embedding customer's transactions, and clear preprocessing guidelines are missing. This paper reviews the recent literature of neural embedding for customer behavior and brings three main contributions. First, we provide a set of guidelines for preprocessing and modeling consumer transaction data to learn neural network embeddings. Second, it is introduced a multi-task Long Short-Term Memory Network to evaluate the guidelines proposed through the task of purchase behavior prediction. Third, we present a multi-contextual visualization of customer behavior embeddings, and its usefulness for purchase prediction and fraud detection applications. Results achieved illustrate accuracies above 40%, 60%, and 80% for predicting the next days, hours, and products purchased for some customers in a dataset composed of online grocery shopping transactions.