When the Umpire is also a Player: Bias in Private Label Product Recommendations on E-commerce Marketplaces

摘要: Algorithmic recommendations mediate interactions between millions of customers and products (in turn, their producers sellers) on large e-commerce marketplaces like Amazon. In recent years, the sellers have raised concerns about fairness black-box recommendation algorithms deployed these marketplaces. Many complaints are centered around biasing to preferentially favor own 'private label' over competitors. These exacerbated as increasingly de-emphasize or replace 'organic' with ad-driven 'sponsored' recommendations, which include private labels. While been covered in popular press spawned regulatory investigations, our knowledge, there has not any public audit marketplace algorithms. this study, we bridge gap by performing an end-to-end systematic related item We propose a network-centric framework quantify compare biases across organic sponsored recommendations. Along number proposed bias measures, find that significantly more biased toward Amazon label compared findings primarily interesting Amazon, measures generally useful for measuring link formation social content networks.