Consumption and Performance: Understanding Longitudinal Dynamics of Recommender Systems via an Agent-Based Simulation Framework

AbstractWe develop a general-purpose agent-based simulation and modeling approach to analyze how user–recommender interactions affect recommender systems in the long run. Our explorations show that, over time, user–recommender interactions consistently lead to the longitudinal performance paradox of recommender systems. In particular, users’ reliance on recommendations, while helping users discover relevant items, actually hurts the future diversity of items that are recommended and consumed as well as slows down the system’s learning pace (i.e., the rate of predictive accuracy improvement). We also demonstrate unique benefits of certain hybrid consumption strategies—that is, that take advantage of both popularity- and personalization-based recommendations—in facilitating improvements in consumption relevance over time. Because users’ consumption strategies can significantly influence the longitudinal performance of recommender systems, it is important for designers to analyze the histories of a system’s recommendations and users’ choices to infer and understand users’ consumption strategies. This would enable the system to anticipate users’ consumption behavior and strategically adjust the system’s parameters according to its long-term performance objectives.We develop a general agent-based modeling and computational simulation approach to study the impact of various factors on the temporal dynamics of recommender systems’ performance. The proposed agent-based simulation approach allows for comprehensive analysis of longitudinal recommender systems performance under a variety of diverse conditions, which typically is not feasible with live real-world systems. We specifically focus on exploring the product consumption strategies and show that, over time, user–recommender interactions consistently lead to the longitudinal performance paradox of recommender systems. In particular, users’ reliance on the system’s recommendations to make item choices generally tends to make the recommender system less useful in the long run or, more specifically, negatively impacts the longitudinal dynamics of several important dimensions of recommendation performance. Furthermore, we explore the nuances of the performance paradox via additional explorations of longitudinal dynamics of recommender systems for a variety of user populations and consumption strategies, as well as personalized and nonpersonalized recommendation approaches. One interesting discovery from our exploration is that a certain hybrid consumption strategy—that is, where users rely on a combination of both personalized- and popularity-based recommendations, offers a unique ability to substantially improve consumption relevance over time. In other words, for such hybrid consumption settings, recommendation algorithms facilitate the general “quality-rises-to-the-top” phenomenon, which is not present in the pure popularity-based consumption. In addition to discussing a number of interesting performance patterns, the paper also analyzes and provides insights into the underlying factors that drive such patterns. Our findings have significant implications for the design and implementation of recommender systems.