RIETI Report May 2021

Machine learning as a natural experiment: Application to fashion e-commerce

The increasing prevalence of machine learning in all areas of industry and public life is adding incredible decision-making capacity to our businesses and institutions as data increasingly drives our understanding of the world. An interesting side-effect of the way that many machine learning algorithms operate is the presence and creation of natural experiments, which can be used for various forms of analysis. In this column, Narita, Aihara et. al. utilize their method to improve the click-through rate of the Japanese e-commerce platform ZOZOTOWN.

This month's featured article

Machine learning as a natural experiment: Application to fashion e-commerce

NARITA YusukeVisiting Fellow, RIETI

AIHARA ShunsukeCo-founder, Hanjuku-Kaso, Inc.; Endowed Assistant Professor, Jichi Medical University

MATSUTANI MegumiDirector, ZOZO Research

SAITO YutaUndergraduate student, Tokyo Institute of Technology

Machine learning algorithms are increasingly being used in decision making. Web companies, car-sharing services, and courts rely on algorithms to supply content, set prices, and estimate recidivism rates. This column introduces a method for predicting counterfactual performance of new algorithms using data from older algorithms as a natural experiment. When applied to a fashion e-commerce service, the method increases the click through rate and improved the recommendations algorithm.

Decision making using prediction by machine learning (ML) algorithms is becoming increasingly widespread (Athey and Imbens 2017, Mullainathan and Spiess 2017). For instance, Amazon, Facebook, Google, Microsoft, Netflix, and other web companies apply machine learning to problems such as personalising ads and content (movies, music, news, etc.), determining prices, and ranking search results. The prices set by car sharing services such as Uber, Lyft, and DiDi are also based on proprietary algorithms based on information about supply and demand at each point in time and location (Cohen et al. 2016).

To read the full text

Recent articles on RIETI's website

Event Information

For a complete list of past and upcoming event information.



BBL Seminars

Fellow titles and links in the text are as of the date of publication.

For questions or comments regarding RIETI Report, please contact the editor.

*If the "Send by mailer" button does not work, please copy the address into your email "send to" field and connect the prefix and the suffix of the address with an "@", sending it normally.

RIETI Report is published bi-weekly.