Algorithmic Governance of Two-Sided Platforms: The Case of Short Video Recommendation

Two-sided platforms face a number of governance challenges such as encouraging participation, managing activities, and fostering network externality. Traditionally, platform governance primarily relies on economic incentives and mechanisms. In this paper, we instead propose an algorithmic perspective to platform governance, and study how important governance objectives can be achieved by changing the recommendation strategy. Taking short video platforms as the specific context, we design an integer programming framework that determines how videos created by producers are recommended to consumers. Different from the common "preference-maximizing" strategy that recommends content solely based on consumer preference, our proposed integer program aims to maximize platform revenue (from ads embedded in videos) subject to constraints that control exposure disparity among producers as well as violations of consumer preferences. We conduct both backtesting and prediction-based evaluations on a public dataset from KuaiShou (a leading short video platform in China) and compare our approach with the preference-maximizing recommendation strategy. We find that recommendations generated by our integer program simultaneously lead to higher platform revenue, lower producer exposure disparity, and improved consumer experiences (in terms of higher information diversity and less repetitive viewing). Further analyses reveal interesting relationships between settings of the integer program and governance outcomes. As an example, we observe that adhering to consumer preferences too much or too little can both increase exposure disparity among producers, and a moderate constraint on consumer preference compliance may sometimes lead to greater producer exposure fairness. Our work contributes a novel algorithmic perspective to platform governance, and offers practical design artifacts that are readily applicable for short video platforms.