Modeling User Behavior Online: A Comparison of Pareto/NBD and BG/NBD Models

Abstract

This study employs user activity data from Steam, which is the largest PC gaming website globally, to compare the effectiveness of two commonly used probabilistic models, the Pareto/NBD and BG/NBD. The primary aim of the research is to evaluate online search engines and assess the vitality of online communities by developing models that simulate customer behavior and calculate the likelihood of customer engagement. The study examines the assumptions and benefits of both models and their limitations. To collect data, user activity data for a six-month period, from January to June 2022, was collected, including user ID, game title, behavior name, and value. The study focuses on the behaviors of "Purchase" and "play," and the value represents the intensity of each action. The research then evaluates the effectiveness of the two models in predicting user behavior, highlighting their strengths and weaknesses. The study concludes by recommending modifications to the BG/NBD model to enhance its accuracy and overcome its limitations. In summary, the study provides valuable insights into the potential applications of the two models and their strengths and limitations in predicting user behavior in online communities.