Research


with Tong Guo and Daniel Yi Xu, Major revision at Marketing Science

Abstract. We study how social media publicity affects the early-stage adoption of sustainable food technologies by local businesses.  Understanding this connection is challenging because of the lack of empirical measurement of local business decisions at scale and, more importantly, the endogeneity of social media publicity to unobserved local demand shocks. Focusing on the case of impossible meat products, we devise a unique location-specific adoption metric based on social media announcements between 2015 and 2019. We propose a novel identification strategy that leverages the quasi-random variations of county-quarter-level news production for different topics to causally identify the linkage between social media publicity and adoption. We find that local news coverage of sustainable food technology increases the adoption of impossible meat products by local restaurants and stores. Interestingly, the elasticity of adoption with respect to news is higher among more liberal regions, and news content about producer financials was the main contributor to identifying the causal effect of news on local businesses' adoption of impossible meat products.


with Yiting Deng and Carl Mela

Abstract. In many digital contexts such as online news and e-tailing with many new users and items, recommendation systems face several challenges: i) how to make initial recommendations to users with little or no response history (i.e., cold-start problem), ii) how to learn user preferences on items (test and learn), and iii) how to scale across many users and items with myriad demographics and attributes. While many recommendation systems accommodate aspects of these challenges, few if any address all. This paper introduces a Collaborative Filtering (CF) Multi-armed Bandit (B) with Attributes (A) recommendation system (CFB-A) to jointly accommodate all of these considerations. Empirical applications, including an offline test on MovieLens data, synthetic data simulations, and an online grocery experiment, indicate that the CFB-A leads to substantial improvement in cumulative average rewards (e.g., total money or times spent, clicks, purchased quantities, average ratings, etc.) relative to the most powerful extant baseline methods. CFB-A's robustness to nonrepresentative training data further highlights its potential to mitigate algorithmic bias across different user groups.

Work in Progress


* Denotes student coauthor at the paper initiation.