主题：Popularity-based or A.I. Personalization-based Algorithms for the Sharing Economy Platform?（共享经济平台） Evidence from Natural Experimentation and Machine Learning（机器学习）
Abstract:This paper examines how platform recommendation algorithms based on demand-side tastes affect supply-side small scale entrepreneurs. It addresses three vital questions: （1） What is the effect of implementations of popularity or AI personalization platform recommendation algorithms on seller revenue? （2） How do changes of platform algorithms incentivize sellers in the sharing economy? （3） For which sellers the platform algorithms are more beneficial? Through natural quasi-experiments and rich proprietary datasets from a major food-sharing platform, the analysis finds significant increases of seller revenues after the platform implements either algorithm. But, the pathways to these revenue increases differ. As the review popularity recommendation （RPR） platform algorithm helps buyers to find sellers with high review ratings more easily, sellers are incentivized to adopt a specialization focus on the quality reputation of current products. By contrast, as the AI bolter personalization recommendation （BPR） algorithm enables buyers to find sellers with more customized cuisines, sellers respond by adopting an innovation focus on introducing more new products to suit the diverse customer tastes. Consistent with the specialization pathway, RPR is more beneficial for sellers who have a concentrated product assortment. In contrast and in line with the innovation pathway, it is younger and newer entrepreneurs that reap more benefits from the BPR. Surprisingly, each algorithm has unintended outcomes: RPR impedes innovation and BPR inhibits specialization. However, the platform manager can leverage a machine learning causal forest technique to learn sellers’ heterogeneous responses to RPR and BPR and craft an optimal targeting rule, which maximizes algorithms’ benefits and minimizes their negative effects for the sharing platform.