Text classification and prediction of purchase product categories on shopping receipts

Recent advancements in mobile devices and campaign management platforms have sparked considerable interest in the realm of personalization concepts in marketing. The emergence of personalized experiences has intensified the competition among retailers within their disruptive milieu. In this competitive environment, presenting accurate personalized offers to the right customers is crucial for maintaining a competitive edge. This study proposes a text classification model and a subsequent purchase behavior prediction model that utilizes the output of the first model as input. The first model predicts the category of purchased products on shopping receipts, while the second model employs individual-level information such as recency, frequency, and spending amount to predict the likelihood of purchasing product categories. Additionally, customers are clustered using the k-means algorithm based on shopping information to investigate potential relationships between clusters and product categories. The performance of tree-based algorithms including decision tree, random forest, gradient boosting trees, is analyzed, and model performances are evaluated using a dataset gathered from a campaign management platform. The results of these experiments indicate that the random forest model outperforms other models with respect to selected performance metrics. Furthermore, in most of the purchased categories, medium effect size correlation is found between purchasing probability of categories and clusters.