The study aims to investigate how recommender systems shape providers’ dynamics and content offerings on platforms, and to provide insights into algorithm designs for achieving better outcomes in platform design.
This study employs a multi-agent simulation framework coupled with reinforcement learning models to examine the influence of different recommender system designs on providers’ perception of demand and platform content.
The study reveals that recommender systems have the potential to introduce biases in providers’ understanding of user preferences, thereby impacting the variety of offerings on platforms. Moreover, it identifies algorithm design as a critical factor, with item-based collaborative filters showcasing superior performance in contexts where customers exhibit selectivity. Conversely, user-based models prove more effective in scenarios where recommendations significantly sway user decisions, ultimately boosting sales.
In practical terms, these insights can guide platform developers in making informed decisions regarding the selection and implementation of recommender system algorithms. By tailoring algorithm choices to specific contexts, platforms can enhance user welfare, ultimately leading to improved platform performance and profitability.
The findings underscore the importance of integrating provider dynamics and algorithmic biases into the design of recommender systems and platforms. This highlights avenues for future research to explore and refine our understanding of these dynamics.
