题　目：News Consumption, Recommender Systems, and Polarization

嘉　宾：Mingduo Zhao (赵鸣铎), Ph.D. Candidate, University of California, Berkeley

主持人：左思 助理教授 上海交通大学安泰经济与管理学院

时　间：2025年9月23日（周二）9:00-10:30

地　点：上海交通大学 徐汇校区 安泰经济与管理学院B404

内容简介:

Recommender systems shape how people consume news, possibly reinforcing political polarization. We run two field experiments to identify how user preferences and algorithms interact to amplify partisan news consumption. In the first study, 2,065 U.S. participants use blank Google accounts and a browser extension to track users' activities on Google News. The first-round recommendations are exogenous, allowing us to show that ideologically aligned content draws more clicks. A second experiment uses bots to randomly click on articles, revealing that each click leads to more aligned content. These two pieces of causal evidence establish a feedback loop between user preference and algorithmic recommendations. We also find in the field study that, after interacting with the recommender system, people's level of polarization increases. A structural model combining a discrete choice model (demand side) with a multi-armed bandit algorithm (supply side) confirms this positive-feedback mechanism. The model is then used to simulate a counterfactual "ideology-blind" recommendation policy that ignores political slant when curating content. While this policy reduces polarization, it comes at the cost of likely lower engagement. Overall, the findings provide causal evidence that personalized algorithms reinforce partisan consumption and exacerbate polarization. They also uncover a fundamental trade-off between mitigating polarization and sustaining engagement, which offers important insights for both platform owners and policymakers.

演讲人简介： 

Mingduo Zhao is a Ph.D. candidate in Economics at the University of California, Berkeley. During Mingduo's Ph.D. studies, Mingduo also completed an M.Sc. in Computer Science and an M.A. in Statistics at UC Berkeley. Prior to joining Berkeley, Mingduo earned a B.Sc. with honors in Mathematics, Statistics, and Economics from the University of Michigan.

His research focuses on how technological advancements profoundly transform platforms and society. With a background in machine learning, causal inference, and empirical industrial organization, Mingduo aims to develop managerial and policy solutions that harness the full potential of emerging technologies while addressing and mitigating their potential risks.

欢迎广大师生参加!