Fuzzy Approach to Purchase Intent Modeling Based on User Tracking For E-commerce Recommenders

Recommender systems play a vital role in ecommerce by presenting personalized product suggestions, reducing habituation and leading to transactions in an environment with limited human touch. Data used for learning how to select optimal recommendation content, including mouse tracking data, are often imprecise in nature. In this paper, we present a fuzzy approach to model purchase intent based on tracking user interaction with a browser via mouse and keyboard. It appreciates data uncertainty and provides insights into e-commerce customer behavior and the development of shops online. The developed fuzzy rule-based systems had a good accuracy and low interpretability, and results show that to generalize possible purchase intent with fuzzy rules, it is good to begin with looking at such behavioral features as distance of mouse movement, distance of vertical page scrolling, number of mouse clicks and time of user activity on website in relation to page content length. In future work, we intend to look at more features reflecting product parameters and transactions to enhance the modeling results on a larger scale.