Abstract:
|
Recommender systems are ubiquitous in online platforms for suggesting items to buy, content to view, people to connect with, etc. A recommender system has two sides: a producer side (e.g., sellers in the marketplace, content creators in feeds, service providers in the sharing economy platform) and a consumer side (e.g., buyers, content viewers, guests). Consumer-side experiments are easy to design and widely used in practice to measure how ranking changes in a recommender impact the behavior of consumers. On the other hand, designing producer-side experiments is challenging because producer items in the treatment and control groups need to be ranked by different models and then merged into a single ranking for each consumer. In this paper, we will develop principles and optimal experiment methods for measuring the effects of ranking changes on the producers.
|