The popularity of e-commerce is increasing day-by-day. In order to provide a seamless experience and tailored offer for the customers, the knowledge of their preferences and behavior is required. The demography of customers is one of the important information used for, e.g., segmentation. To provide optimal service, machine learning is often used for various tasks. However, the amount of data generated by customers and also large and changing product catalogs result in poorly performing models. In this paper we aim at introducing behavior-based abstraction, which includes item and event abstractions also. Our method reduces the number of unique items in the e-commerce catalog and in the next step encodes the user behavior. We performed extensive evaluation over the real-world dataset. The results suggest the usefulness of proposed abstraction and also resulted in the improvement of the demography prediction performance both from model complexity and performance point of view.