Product ranking, pricing, and recommendation for e-commerce retailers and platforms

Product ranking and pricing in an online store or marketplace plays a crucial role in customer engagement and subsequently in sales and total revenue. An online retailer or marketplace operator must decide on the set of the products and their positions and prices at various points, such as campaigns, first page, landing pages, side-bar recommendations, or search query lists. In this dissertation, we investigate the optimal ranking and pricing in centralized and decentralized environments at online e-commerce platforms. In the first part, we propose a new personalized content-based method specially designed for online retailers. The focus of the study is on customers' activities in an online store, such as clicks, and purchases, more than the feedback such as ratings or comments. This property is useful for online retailers where there is limited rating feedback, but there are several search sessions including clicks and purchases. We test the method using data provided by an apparel retailer. Our method outperforms benchmark methods (Collaborative filtering, Popular products, and MNL choice model) and it has a strictly better performance in recommending new products. Also, our method outperforms benchmark methods in recommending products to customers who are generally interested in less popular (fringe) products. The main practical contribution of this study for online retailers is a novel personalized recommending system that is build-up based on panel data (instead of rating data) that causes better use of limited detail provided in the dataset and improved the quality of recommendation. In the second part, we consider a decentralized pricing and ranking problem in an online marketplace where sellers decide on their prices simultaneously and ranking takes place based on a pre-defined rule by the marketplace operator (MO). Empirical studies show that the ranking of sellers in an online marketplace is highly correlated to prices. We model the price competition as a full information game where all players (sellers) know the price-based ranking rule defined by the marketplace operator and the demand function of each seller. We obtain the Nash equilibrium of each game under binary and probabilistic ranking policies and show how ranking policy and commission rate affect the pricing strategies and equilibrium. Then, we investigate the profit of MO and third-party sellers dependent on the ranking rules and market structure. From a practical and implementation point of view, this study sheds light on the optimal pricing and ranking strategies in an online marketplace with three main contributions. First, we find the optimal ranking rule for MO is dependent on the sale margin and cost of the products. Further, we show that the MO may deviate from the optimal ranking to either increase the profit of each supply-side party by implementing a less competitive ranking rule and higher commission rate or enhance the consumers' experience quality such as demand satisfaction rate and demand surplus by implementing a competitive ranking rule and lower commission rate. Finally, in cases that MO participates as a seller, s/he may rank himself higher than other sellers. This strategy increases the profit of MO in any case, however, surprisingly it increases both total supply-side profit and customers demand surplus if the demand functions are asymmetric. In the third part, we study product ranking and pricing at e-commerce platforms, where customers are generally window shoppers. Window shoppers are a type of customer that browse online stores as a form of leisure or external search behavior without a clear intent to purchase a specific product. We develop an optimization model with deterministic demand and vertically differentiated positions (ranking) where products that are placed in earlier positions are more likely to be considered and, therefore, purchased. We show that revenue-ordered ranking is optimal when there is infinite inventory for products. However, when there is a finite inventory case, despite the deterministic and invertible structure of demand, we prove that the optimal price would result in product stock-out during the campaign in some cases. Also, we show that solving the problem with a set of predefined prices (and not joint pricing ranking) would result in insignificant profit loss if the diversity of the product's demand parameters is limited. The precise optimization model in the infinite inventory case is intractable due to the existence of integer variables and the complexity of the constraints. Thus, we propose two heuristic methods to solve the problem and provide an upper bound by relaxing the integer variables. We show that the performance of the heuristic methods is dependent on the customer's search behavior. Selecting the appropriate heuristic method, we provide a near-optimal solution in all cases. Finally, numerical analyses show that trivial solutions such as sorting by price or sorting by inventory would result in significant profit loss.