Assortment optimization under cardinality effects and novelty for unequal profit margin items

This work focuses on assortment optimization problems concerned with the three ways a recommender system increases conversion rates: (1) improve accuracy, (2) provide consideration sets and (3) introduce novelty and diversity. To do so, we introduce a new random utility-based model, which in addition to item and user attributes, captures context effects, as well as the need to introduce novel items. We use this new model and an existing random utility model to study assortment optimization problems in which the value that a customer derives from an assortment reaches a maximum and then begins to decline with increase in size. We focus on an online retail environment, in which a trade-off exists between recommending items with high profit margins, but low consumer choice probability. We present polynomial-time algorithms to solve both optimization models. We computationally analyze the performance of the algorithms using real-world online transactional data.