Journal Papers under Review

Longitudinal Impact of Preference Biases on Recommender Systems' Performance, with Jingjing Zhang, and Gedas Adomavicius. [PDF]
Click for Abstract

Research studies have shown that recommender systems' predictions that are observed by users can cause biases in users' post-consumption preference ratings. Because users' preference ratings are typically fed back to the system as training data for future predictions, this process is likely to influence the performance of the system in the long run. We use a simulation approach to study the longitudinal impact of preference biases (and their magnitude) on the dynamics of recommender systems' performance. We look at the influence of preference biases in two conditions: (i) during the normal system use, where biases are typically caused by the system's inherent prediction errors, and (ii) in the presence of external (deliberate) recommendation perturbations. Our simulation results show that preference biases significantly impair the system's prediction performance (i.e., prediction accuracy) as well as users' consumption outcomes (i.e., consumption relevance and diversity) over time. The impact is non-linear to the size of the bias, i.e., large bias causes disproportionately large negative effects. Also, items that are less popular and less distinctive (in terms of their content) are affected more by preference biases. Additionally, intentional recommendation perturbations, even on a small number of items for a short time, substantially amplify the negative impact of preference bias on a system's longitudinal dynamics and causes long-lasting effects on users' consumption. Our findings provide important implications for the design of recommender systems.

Major Revision at Information Systems Research

Healthcare across Boundaries: Urban-Rural Differences in the Financial and Healthcare Consequences of Telehealth Adoption, with Xuelin Li, and Gordon Burtch. [PDF]
Click for Abstract

We study the impacts of telehealth adoption on geographic competition among urban and rural healthcare providers, and associated quality of care implications. To causally identify these effects, we consider a quasi-natural experiment: states' entry into the Telemedicine Licensure Compact, wherein participating states coordinate to streamline licensing for physicians wishing to provide telehealth services across state lines. We first show that affected physicians receive more state licenses and earn higher Medicare payments, thereby establishing the Compact entry shock's validity and its positive effect on telehealth adoption. We then examine the heterogeneous effects on provider earnings and quality of care across urban and rural areas. We report evidence that urban providers are systematically more likely to respond to the policy change and financially benefit from it by expanding their service scope to a wider geographic market. As urban providers begin to offer their services to rural patients, rural physicians and hospitals experience a decline in patient volumes, and a revenue loss in turn. We subsequently consider parallel impacts on patient quality of care, and we discuss the implications of our results for healthcare providers and government.

Under Review at Information Systems Research

Short-Lived Item Recommendation, with Yicheng Song, Zhuoxin Li, and Chunmian Ge. [PDF]
Click for Abstract

In highly dynamic markets (e.g., online news, short videos, and flash sales), new items continuously flow into the markets and fade out quickly. Also, users' preferences shift as they interact with a fast-evolving item pool. These two characteristics of short-lived items make personalized recommendation an essential but challenging task. The absence of user-item interactions for new items brings in the cold-start problem. Recommender systems also need to promptly update user and item representations to incorporate new interactions, so that the systems immediately yield improved recommendations for users throughout the short life-circle of items. To address these challenges, we propose Attention Initialized Dual Recurrent Neural Network (AI-DRNN), a holistic deep learning model that learns effective initial representations of new items via Attention Mechanism, and efficiently updates users/items representations via Dual Recurrent Neural Network to improve recommendation accuracy. We evaluate the proposed model with clickstream data from an online flash sale platform and report its performance in predicting consumer behavior regarding which product a consumer will interact with next and the type of interaction (placing an order or continuing to search). Empirical experiments show that the proposed AI-DRNN performs significantly better in prediction accuracy on cold-started items and overall, compared with the state-of-the-art benchmarks. We further explore the item representations to understand the dynamic updating process of AI-DRNN. We find that AI-DRNN demonstrates interpretability regarding how the algorithm works. Interestingly, we find the proposed AI-DRNN shares the key property of collaborative filtering theory (i.e., predicting a user’s interests by pooling preference information from other users).

Under Review at Information Systems Research

Working in Progress

Economic Impact of Category-Expansion-Oriented Recommendations: Evidence from Randomized Field Experiments, with Ravi Bapna, Gedas Adomavicius, and Jonathan Hershaff.

Impact of Data Privacy Regulations on Recommender Systems, with Yicheng Song, and Gedas Adomavicius.

Self-Consistency of Machine Learning Techniques, with Gedas Adomavicius, and Jingjing Zhang.