Integrative models and empirical analysis of recommender systems in online retailing

Date of Completion

January 2006


Business Administration, Marketing|Business Administration, Management|Information Science




Online retailers are increasingly utilizing recommender systems to offer product recommendations to consumers. Such recommendations are typically based on previous purchases made by a network of customers with related purchase patterns. Although there has been extensive research devoted to enhancing the quality of recommendations, little research has been done in integrating recommendations with economic factors that drive the purchase behavior. Moreover, much of the work to date has utilized data collected from user satisfaction surveys or simulated experiments to assess the impact of recommender systems. This dissertation adds to the growing literature in recommender systems by providing models to integrate other economic factors along with recommendations and by empirically investigating the performance of online recommendations. ^ The first essay of this dissertation presents integer programming based models to integrate recommendations with economic factors related to consumer purchase behavior. The underlying contention is that even if recommendations are accurate and useful, customers may not be interested in purchasing if such recommendations are not properly aligned with their economic interests. Integer programming models developed in this essay deal with integrating various economic incentives such as online retail promotions and price discounts with recommendations. These models are complementary to current recommender system algorithms and can be implemented by online retailers, or, by other independent intermediaries such as shopbots. The empirical analysis for these models suggests that our alternative set of recommendations offer significantly higher economic benefits to customers. ^ The second essay empirically analyzes the relationship between sales and recommendations. The analysis is based on a panel data of books collected from publicly available information from online retailers. A weighted measure for recommendations is developed based on the number and the impact of recommenders. Subsequently, pooled OLS and panel data based models are used to analyze the effect of recommendations on sales. Further, sales, price and recommendations are jointly determined using a simultaneous equations model to address the potential endogeneity arising from simultaneity amongst these variables. We estimate the marginal change in sales due to recommendations, and contrast it with the impacts of other types of customer feedback.^